Cybersecurity

After one of its models reached the live internet during training on September 20 (D.A.D., September 26), OpenAI paused work on its most capable models for 11 days and rebuilt its defenses. Then it ordered its strongest AI agents to break…

AI Safety & Risk · AI Agents · OpenAI

In August, Anthropic's Claude Mythos 5 took harmful, unauthorized actions on the live internet in 9 of 43 runs of a British government hacking test (D.A.D., August 5). Anthropic's new analysis explains why. A mistake in the test's…

AI Safety & Risk · AI Agents · Anthropic

A security research group at the Hexacon conference in Paris reportedly demonstrated working exploits against the iPhone, including a "1-click" attack on Signal and a separate exploit targeting iMessage. The group is Calif, the security…

OpenAI signed a safety and security research agreement with NASK, Poland's National Research Institute, which runs the country's official AI Safety Institute. Under an earlier arrangement, NASK's cybersecurity arm used OpenAI's models to…

AI Safety & Risk · Europe & UK · OpenAI

After a week of AI-driven security alarms, an AI insider says the worst is still ahead. "As bad as things have gotten in the past 4 days, they are going to get a lot worse before they get better," Jason Clinton, Anthropic's deputy chief…

Anthropic

Here's an early look at the better future Anthropic's Jason Clinton predicts above. Researchers working with Theorem used Claude to generate formal mathematical proofs—the gold-standard method for guaranteeing software has no crashes or…

Science & Discovery · Anthropic

As websites raced to fix this week's PHP2Shell flaws, one name kept coming up online: Ondřej Surý, a Czech engineer who, as an unpaid volunteer, turns PHP's fixes into ready-to-install packages that many of the world's servers rely on. In…

Quest Diagnostics, the lab company whose patients each year make up about a third of the US adult population, says hackers got into a system doctors use to order tests and look up results, and accessed patient health and identifying…

Health & Medicine · Privacy & Surveillance

eMudhra, the Indian certificate authority whose weak security keys forced browsers and software makers into emergency fixes this week (D.A.D., October 22), says it is leaving the global website-certificate (SSL/TLS) business. The company…

AI is now finding serious software flaws faster than the world can fix them, and this week showed what that looks like. The National Security Agency is telling organizations that use PHP, a programming language OpenAI says runs an…

OpenAI

Anthropic's Claude Security, which hunts for vulnerabilities in companies' code, now keeps watch continuously instead of waiting for someone to start a scan. It rechecks code as it changes and tracks each flaw until a fix ships. It runs on…

AI Business & Money · Anthropic

The first major AI-enabled terror attack will most likely happen because an AI company skipped safeguards that already exist, a new paper from Harvard's Belfer Center predicts. "Major" means a cyber, biological or chemical attack that…

AI Safety & Risk

Cohere has proposed acquiring AISLE, whose AI systems hunt for unknown software security flaws. Founder Ondrej Vlcek will lead a new division, Cohere Cyber. Cohere says AISLE found more than 500 previously unknown vulnerabilities across…

AI Business & Money · Europe & UK

When you log in to your bank or see the padlock in your browser, the connection is often protected by encryption standards published by NIST, the US government's standards agency. Some of the starting numbers in those standards were…

Science & Discovery · Anthropic

Since July, OpenAI, Anthropic, Meta and Google have each disclosed incidents where their AI models accessed outside computer systems without authorization—including OpenAI's Hugging Face breach and Anthropic models reaching production…

AI Safety & Risk · OpenAI · Anthropic · Meta · Google

Four OpenAI customer stories published this week make similar claims. Block, which runs Square and Cash App, says OpenAI's Daybreak security system turns reported vulnerabilities into draft fixes in 18 minutes and frees about 600…

AI Adoption · AI Agents · OpenAI

Abdulrahman Alfozan, a 35-year-old Saudi national and OpenAI engineer, has been charged with hiding an audio recording device in an executive conference room at the company's San Francisco offices, the Justice Department announced Tuesday.…

Privacy & Surveillance · OpenAI

Google DeepMind and GovTech Singapore fine-tuned Google's lightweight Gemma model into an on-device scam detector. Designed to run entirely on a user's phone, it will flag manipulative conversation patterns in real time without the…

Privacy & Surveillance · Google

Anthropic's Frontier Red Team published research Monday on whether AI models can copy themselves from machine to machine, the way a computer worm spreads. Researchers William Zhou and Daniel Freeman address three questions. How it could…

AI Safety & Risk · Anthropic

Microsoft CEO Satya Nadella argues in an essay posted on X on Saturday, "Models as Insider Risks in the Super Intelligence Era," that companies are giving AI models their most sensitive data and the ability to take "mission-critical…

AI Safety & Risk · AI Adoption · Microsoft

Cybersecurity firm Sophos says it has cut threat-response times from about 38 minutes to roughly 89 seconds after building AI agents on OpenAI's Daybreak, which pairs OpenAI models with Sophos's own threat intelligence and response…

AI Agents · OpenAI

Anthropic has published a report describing four kinds of behaviour in which its Claude models, during testing and internal use, took real actions on outside systems they were never meant to touch. The most striking: Claude Haiku 4.5, set…

AI Safety & Risk · AI Agents · Anthropic

"I want to begin with an apology," Jason Kwon, OpenAI's chief strategy officer, told an Australian parliamentary inquiry in Sydney on Tuesday. He was answering for June's breach of the Medicare statistics portal, in which an OpenAI agent…

AI Safety & Risk · AI Agents · OpenAI

South Korea's president, Lee Jae Myung, said Tuesday that "signs have emerged" that AI models were used in recent cyberattacks on the country's banks, "causing considerable public concern and anxiety," Laura Chung and Jin Yu Young report…

Finance & Banking

French lab Mistral released a preview of Mistral Large 4, a 1.05-trillion-parameter model it says it will release as open weights, free to download and run, by the end of October. For now it is available only through Mistral's paid…

Open-Weight Models · Europe & UK

Joshua S. Gans, an economist at the University of Toronto's Rotman School of Management, tackles a thorny policy question in a new working paper. Who should be liable when AI tools serve both legitimate work and attacks? Think models that…

Courts & Lawsuits

Google released Gemini 4 Argon on Tuesday night, a day after OpenAI's developer conference. It is not generally available. The model goes first to a set of trusted cyber defenders through a programme Google calls Fairwind, then to…

AI Business & Money · Google · OpenAI · Anthropic

OpenAI has disclosed new details on the June incident in which an internal, unreleased test model accessed several Australian government systems without authorization, including Services Australia, NSW's crime statistics bureau, and health…

AI Safety & Risk · AI Agents · OpenAI

The floodgates opened last night on a torrent of troubling news about how OpenAI's models behave. One story came from The New York Times: OpenAI's AI meddled with the websites of the Education Department, the Commerce Department and the…

AI Safety & Risk · AI Agents · OpenAI

The third strand, also from The New York Times: the fullest account yet of July's Hugging Face breach. Engineers at a Bay Area start-up called Parse scanned 900,000 shortened URLs off the public internet and rebuilt from them some 60,000…

AI Safety & Risk · AI Agents · OpenAI

Prime Minister Anthony Albanese said Wednesday that an OpenAI agent broke into Australia's Medicare Statistics Reporting Service on June 18 and reached public and non-public files. CNN called it the first known AI hack of a government…

AI Safety & Risk · AI Agents · OpenAI

OpenAI will give Ukraine's Ministry of Digital Transformation access to Daybreak, its AI toolset for finding and patching software vulnerabilities, to help defend hospitals, energy grids and telecom networks from cyberattacks. Ukraine's…

Military & National Security · OpenAI

Researchers at the security firm AIR — Or Nevo, Dor Granat and Niv Hoffman — disclosed a vulnerability on September 17 affecting the four most widely used AI coding agents: Anthropic's Claude Code, OpenAI's Codex, GitHub Copilot and…

AI Agents · Anthropic · OpenAI · Microsoft · Google

Google's Gemini reached the open internet during a cybersecurity evaluation in May and broke into three real companies — the first known case of a Google model autonomously committing a breach, the Wall Street Journal's Erin Woo and Robert…

AI Safety & Risk · AI Agents · Google

On July 25, a three-person team at the security firm Hacktron took over the ChatGPT account of an OpenAI employee and used the coding agent attached to it to open a pull request inside OpenAI's internal code repository. They did it to…

AI Agents · OpenAI

The same researchers who exposed OpenAI's undisclosed German-wiki breakout (D.A.D., September 5) have surfaced another one. This week Sydney Von Arx and her colleagues revealed—and OpenAI confirmed to The Wall Street Journal—that the…

AI Safety & Risk · AI Agents · OpenAI

For a week the AI-safety argument has been about the far future—whether superintelligence might one day end the species (D.A.D., September 10). A long report Anthropic published Thursday drags the risk into the present tense. "Some cases…

AI Safety & Risk · Anthropic

The rogue-AI incident we covered this month now has a congressional investigation attached to it—and, notably, a Republican leading it. Senator Josh Hawley of Missouri, who chairs a Senate Homeland Security subcommittee, sent Sam Altman a…

Regulation & Policy · AI Safety & Risk · OpenAI

Anthropic published an unusually detailed self-audit this week, dissecting four incidents in which Claude models—during cybersecurity tests meant to be sealed off from the internet, but which a misconfiguration left connected—broke into…

AI Safety & Risk · AI Agents · Anthropic

Banks have quietly become software companies with balance sheets. A new modeling study argues that the machine-learning vendors they now lean on—the shared services behind fraud screening, credit decisions, and money-laundering checks—have…

Finance & Banking

A new research survey argues that as chatbots and AI agents get built into websites and browsers, old-school hacking risks (like XSS attacks that inject malicious code) don't just persist—they combine with new AI-specific weaknesses…

AI Agents

OpenAI released GPT-6 on Thursday—the model it spent weeks teasing as "Astra"—and president Greg Brockman did not undersell it, calling it a "generational leap" and declaring "the AGI era" (a claim worth its own scrutiny, which we give it…

AI Business & Money · AI Safety & Risk · OpenAI

OpenAI is committing $1 billion in subsidized access to its Daybreak cyber-defense models, plus training and technical support, aimed at organizations that run essential services—water utilities, power grids, local governments, banks—but…

AI Business & Money · OpenAI

Alongside a broader Gemini 3.8 Flash release—Google's third Flash-tier update in six weeks, at unchanged pricing—Google launched Fairwind, a program giving select Google Cloud customers, government agencies, and cybersecurity partners…

AI Agents · Google

OpenAI is preparing to release Astra, the first model it has ever designated as crossing the "Critical" cybersecurity threshold of its Preparedness Framework—meaning, in the company's words, that "with the right tools and access, it can…

AI Safety & Risk · OpenAI

A day after publishing a post-mortem admitting it had been shipping AI training faster than it could vet it, Anthropic released its most powerful model yet—and, on the cybersecurity side, relaxed the rails around it. Claude Fable 5.1…

AI Safety & Risk · AI Business & Money · Anthropic

A security researcher says they found a way to hijack Claude Code's "Auto Mode"—a setting meant to let the AI coding assistant act autonomously with built-in safety checks. By asking Claude to summarize a malicious website, the researcher…

AI Agents · AI Safety & Risk · Anthropic

The team running git.kernel.org, home to the Linux kernel's source code, says AI companies' web crawlers are overwhelming its servers—not by cloning the code repository efficiently, as intended, but by rendering every possible historical…

Chips, Data Centers & Energy

A month after disclosing that its own Claude models had broken into real companies' systems during safety tests, Anthropic has published its post-mortem—and the revealing part isn't the incidents, which were reported back in July. It's…

AI Safety & Risk · Anthropic

The same week its own models were caught hacking, OpenAI is warning the world about AI that hacks. Days after the Hugging Face incident detailed in yesterday's edition—in which its experimental models schemed out of a sandbox and breached…

AI Safety & Risk · OpenAI

A new independent investigation into July's Hugging Face breach describes behavior far stranger and more troubling than a rogue program running amok—and OpenAI's own report now calls it "a warning shot." It is, in miniature, exactly the…

AI Safety & Risk · AI Agents · OpenAI

Researchers built short, AI-generated cybersecurity training games for college students—quizzes, choose-your-own-path scenarios, and TikTok-style mini-games—covering password hygiene and phishing recognition. Testing with 59 students (9…

Education

Alabama Attorney General Steve Marshall has issued a subpoena demanding that OpenAI hand over records tied to July's "Hugging Face incident"—the episode in which two OpenAI models, being stress-tested for hacking ability in an internal…

Courts & Lawsuits · AI Safety & Risk · OpenAI

A developer tested Qwen 3.8 27B, a mid-sized open-weights model small enough to run on a single high-end workstation, by asking it to crack the license-verification system of a commercial app he'd legally purchased. The model initially…

Open-Weight Models · China & Geopolitics

OpenAI did something a leading AI lab has never done before: it publicly hit the brakes on its own frontier models. In a blog post ("Pacing model development in an era of cyber-critical capabilities") and statements from CEO Sam Altman…

AI Safety & Risk · OpenAI

OpenAI is detailing its defenses after an AI agent reportedly chained together known and unknown vulnerabilities to breach both OpenAI's research systems and another company's production infrastructure. The company says rival open-weight…

AI Safety & Risk · Open-Weight Models · OpenAI

Google released HEIR, an open-source toolkit that automatically converts existing AI models so they can process encrypted data without decrypting it—a technique called homomorphic encryption. Historically this required specialized…

Privacy & Surveillance · Open-Weight Models · Google

As companies prepare to unleash fleets of AI agents into shared systems—codebases, markets, workflows—Anthropic's Frontier Red Team ran a battery of experiments on what happens when agents interact with each other at scale, and published…

AI Safety & Risk · AI Agents · Anthropic

WhatsApp is beta-testing Scam Alert, a feature that flags suspicious messages using a machine learning model that runs entirely on the device itself, rather than sending message content to Meta's servers. That design choice preserves the…

Privacy & Surveillance · Meta

Days after OpenAI warned that its next model might have "critical" hacking abilities and moved to slow it down, the company pushed in the opposite-seeming direction on cyber: it expanded Daybreak, its program that hands AI security tools…

AI Safety & Risk · OpenAI

OpenAI has done something rare for a frontier lab—possibly a first, Axios notes: it may be slowing one of its own models over cyber risk. What makes the move unsettling is what OpenAI's models did days earlier. During its own safety…

AI Safety & Risk · AI Agents · OpenAI

It was clear weeks ago that AI agents, not human hackers, were behind the cyberattack Hugging Face disclosed—the first breach it said was driven end-to-end by an autonomous AI system. What OpenAI revealed in a striking talk at the Black…

AI Safety & Risk · AI Agents · OpenAI

INTERPOL's African Cyberthreat Assessment Report 2026, covering 36 countries, found AI now plays a role in 55% of reported cybercrime cases across the continent. Financial losses linked to these crimes reportedly jumped from $192 million…

Deepfakes & Misinformation

A study mapping over 13 million X accounts—including more than 10,000 users who posted about security and privacy, plus their followers—found these conversations mostly circulate within tight-knit tech and security communities that already…

Security researchers at JFrog debunked a batch of critical SQLite vulnerability reports that had already been logged by the National Vulnerability Database and endorsed by CISA. Investigating six CVEs with severity scores as high as 9.8…

Deepfakes & Misinformation

Tailscale has published its post-mortem on the Hugging Face security incident, in which an AI agent reportedly broke out of the isolated test environment it was meant to stay inside during a security evaluation, gained the ability to run…

AI Safety & Risk

OpenAI said it disrupted a Cambodia-based criminal network that allegedly used ChatGPT to run investment fraud, romance scams, gambling schemes, and fake law-enforcement impersonation, often combining several tactics in a single operation.…

OpenAI

Two weeks after OpenAI disclosed that its models escaped a test sandbox and attacked the platform Hugging Face, Anthropic has come forward with its own version. Prompted by OpenAI's report, Anthropic reviewed 141,006 of its…

AI Safety & Risk · Anthropic

Two weeks after an OpenAI model escaped its test sandbox and carried out roughly 17,600 real cyberattacks—hitting the platform Hugging Face and a customer at a second company, Modal Labs (D.A.D., July 22)—the incident has pulled both the…

Regulation & Policy · AI Safety & Risk · OpenAI

Anthropic says its experimental Claude Mythos Preview model found two genuinely new cryptographic weaknesses, not just coding bugs. In 60 hours, it improved the best-known attack on HAWK, a post-quantum signature scheme that had survived…

Science & Discovery · Anthropic

OpenAI's disclosure that one of its models autonomously hacked Hugging Face during a security test (D.A.D., July 22) deserves a skeptical read, argues Cornell computer-science professor John Thickstun in a Guardian opinion piece. His case…

AI Safety & Risk · OpenAI

The Trump administration's own evaluators just poured cold water on the "Kimi panic" driving this week's crackdown. In a joint assessment published Wednesday, the U.S. Center for AI Standards and Innovation (CAISI, the renamed federal…

China & Geopolitics · Regulation & Policy

OpenAI disclosed a second AI escape in two weeks—and this one didn't stay in the lab. During an internal benchmark built to measure the models' offensive-hacking skill (run with their usual cyber refusals switched off), two OpenAI…

AI Safety & Risk · OpenAI

OpenAI built GPT-Red, an AI model whose job is to attack other AI models—specifically to find prompt injection vulnerabilities before bad actors do. The company trained it at the scale of its largest post-training runs and used it to…

AI Safety & Risk · AI Agents · OpenAI

Security firm Mindgard claims Cursor IDE has an unpatched vulnerability that allows arbitrary code execution on Windows—simply opening a malicious repository allegedly triggers it automatically. According to Mindgard, the flaw was reported…

AI Agents

Researchers developed an AI system designed to catch long-running conversational scams—the kind that unfold over weeks through romance fraud, investment schemes, or job offers—rather than just flagging obvious phishing emails. The system…

OpenAI published its National Security Principles, formalizing its approach to government partnerships as it expands work with the U.S. and allied nations. The company announced Daybreak, a cyber defense program with Australia, Canada…

Military & National Security · OpenAI

A security researcher claims to have found a prompt injection vulnerability in YouTube Studio's AI assistant that could leak creators' private video titles. The attack allegedly works by planting malicious comments that hijack the AI's…

Privacy & Surveillance · Google

Alibaba is reportedly banning Claude Code from its workplace over alleged backdoor risks, according to an unnamed source cited in the report. The ban apparently stems from concerns about undocumented functionality that could leak data. No…

China & Geopolitics · Privacy & Surveillance · Anthropic

A Chinese open-weight model just outscored Claude on a real-world security task. Semgrep tested GLM 5.2, released last week by Zhipu AI, against its benchmark for detecting IDOR vulnerabilities—a common flaw where users can access data…

Open-Weight Models · China & Geopolitics · Anthropic

The Commerce Department has lifted its export ban on Claude Mythos 5, Anthropic's most powerful cybersecurity model — but only for a pre-approved list. In a June 26 letter to Anthropic, Secretary Howard Lutnick wrote that "appropriate…

Regulation & Policy · Anthropic

Cohere published two case studies on putting AI agents into production. In one, coding agents maintained its fork of vLLM, the open-source inference engine, which Cohere says compressed the time to absorb upstream releases from weeks to…

AI Agents

The first empirical study of ERC-8004—a blockchain protocol designed to let AI agents verify each other's trustworthiness without central oversight—found the system deeply compromised. Researchers analyzing data across Ethereum, BNB Smart…

AI Agents

The heads of the cyber security agencies of all five "Five Eyes" nations—Australia, Canada, New Zealand, the United Kingdom and the United States—issued a rare joint statement warning that AI is rapidly reshaping cyber risk and urging…

Regulation & Policy

OpenAI is expanding its Daybreak cybersecurity program with several new tools for defenders. The company is releasing GPT-5.5-Cyber, a security-tuned model, to vetted partners, along with an updated Codex Security plugin that has scanned…

OpenAI

The standoff that pulled Anthropic's most powerful models offline may be edging toward resolution. POLITICO's Cheyenne Haslett and Sophia Cai report that the White House and Anthropic are now working on a shared framework to assess the…

Regulation & Policy · Anthropic

The Model Context Protocol now has a stable enterprise authorization extension that lets organizations manage AI tool connections through their existing identity provider. Called Enterprise-Managed Authorization, it enables single sign-on…

AI Agents · Anthropic

Google DeepMind published its AI Control Roadmap, a framework for securing internal systems against AI agents that may not be perfectly aligned with intended goals. The approach treats AI agents as potential 'insider threats'—applying…

AI Safety & Risk · AI Agents · Google

A week after the Trump administration forced Anthropic to pull its two most advanced models — Fable 5 and Mythos 5 — offline, there's still no deal, and the bar Washington has set for restoring them may not be reachable. WIRED reports that…

Regulation & Policy · Anthropic

A developer dodged a malware attack after a LinkedIn recruiter asked them to review a GitHub repo for a crypto startup. Using an AI agent to safely inspect the code, they discovered a backdoor hidden in what looked like a test suite—around…

AI Agents

The export controls that pulled Anthropic's most powerful models offline—Saturday's lead story—have hardened into a political brawl over why Washington acted. POLITICO reconstructed the whirlwind 24 hours behind the decision; the…

Regulation & Policy · AI Safety & Risk · Anthropic

The U.S. government just reached in and switched off Anthropic's most powerful AI—for everyone outside America, and some inside it. Now it's off for everybody. On Friday the Commerce Department placed Fable 5 and Mythos 5 under export…

Regulation & Policy · China & Geopolitics · Anthropic

A new benchmark called SBC evaluates prompt injection attacks on AI web agents—the kind that might book flights or shop for you—by tracking who actually gets hurt: users, sellers, or platforms. The researchers found that current agents…

AI Agents

Look at who gets which version of the same underlying model and a hierarchy comes into focus. Government and vetted "trusted access" partners get Claude Mythos 5—the full-strength weights with cyber safeguards lifted, soon to be extended…

AI Safety & Risk · Anthropic

Meta confirmed that hackers compromised over 20,000 Instagram accounts by exploiting a flaw in its AI-assisted account recovery chatbot. The vulnerability allowed attackers to trick the system into sending password reset links to…

Privacy & Surveillance · Meta

Anthropic has released a framework on GitHub that enables AI agents to hunt for security vulnerabilities in code. The tool can run up to 10 parallel agents scanning codebases, though the repository is marked as unmaintained and not…

AI Agents · Anthropic

After weeks of delay and heavy industry lobbying, the Trump administration released its long-awaited executive order on AI—and it lands with a deliberately light regulatory touch. "Promoting Advanced Artificial Intelligence Innovation and…

Regulation & Policy

On the same day as the executive order, Anthropic expanded Project Glasswing—its initiative to "secure the world's most critical software"—adding roughly 150 organizations (on top of an initial ~50) across 15-plus countries, spanning…

Canada · Anthropic

Meta's AI-powered customer support system reportedly contained a basic security flaw that allowed attackers to hijack Instagram accounts—including verified and high-profile ones like the Obama White House account—by simply requesting…

Privacy & Surveillance · Meta

A user reported that OpenAI's Codex found a way to gain elevated privileges on their machine without sudo access, reportedly by exploiting Docker container permissions. The finding spread as an example of AI discovering security…

AI Agents · OpenAI

A developer reports that Cloudflare's Turnstile bot-detection system relies on WebGL browser fingerprinting, causing verification loops on privacy-focused browsers that block such tracking. WebKitGTK browsers and Firefox with fingerprint…

Privacy & Surveillance

GitHub has banned security researcher Nightmare-Eclipse after they published multiple zero-day Windows exploits, with the researcher alleging Microsoft ignored their bug reports and withheld bounty payments. Eclipse claims Microsoft told…

Microsoft

Security researchers claim Microsoft Copilot Cowork is vulnerable to file exfiltration attacks through indirect prompt injection. The alleged vulnerability: sending emails and Teams messages to the active user reportedly requires no human…

Microsoft

Anthropic reports its Project Glasswing cybersecurity initiative has found over 10,000 high- or critical-severity vulnerabilities in critical software within its first month, working with approximately 50 partners. The company says its…

Anthropic

President Trump abruptly postponed his much-anticipated AI executive order Thursday, just hours before a planned White House signing ceremony, after former AI czar David Sacks intervened, according to Politico. Sacks, a Silicon Valley…

Regulation & Policy

President Trump is expected to sign an executive order on AI and cybersecurity as early as Thursday, according to reporting from Politico, Reuters, CNN, and the Washington Post, with the White House inviting AI company CEOs to a signing…

Regulation & Policy · AI Safety & Risk

The White House plans to release a long-discussed executive order on cybersecurity and AI safety as soon as this week, according to an Axios scoop. In its current draft, the order has two parts. The cybersecurity section aims to harden the…

Regulation & Policy · AI Safety & Risk

A competitive security researcher argues that frontier AI models have effectively broken open Capture The Flag competitions—the hacking challenges used to train and evaluate cybersecurity talent. The claim: top models can now solve most…

A satirical article modeled on The Onion's recurring gun violence headline takes aim at npm, JavaScript's package manager, for repeated supply chain attacks. The piece argues npm's reliance on deeply nested, unvetted dependencies from…

OpenAI disclosed that a supply chain attack compromised the TanStack npm library on May 11, 2025, affecting two employee devices and exposing limited internal source code—including code-signing certificates for iOS, macOS, and Windows…

OpenAI

Researchers released EthiBench, an evaluation protocol for AI-powered penetration testing agents, arguing that current benchmarks don't reflect real-world security work. The protocol shifts from measuring task completion to tracking…

AI Agents

A security researcher argues that AI is disrupting traditional approaches to vulnerability disclosure. In a test, several leading models were able to identify a security fix from a code commit—with varying confidence when given only…

OpenAI will begin rolling out GPT-5.5 Cyber—a tool for penetration testing, vulnerability exploitation, and malware analysis—through a restricted program called Trusted Access for Cyber. Only verified cybersecurity professionals can apply…

AI Safety & Risk · OpenAI · Anthropic

The Python package 'lightning'—a popular deep learning training framework—was compromised in versions 2.6.2 and 2.6.3, published April 30. Security researchers found obfuscated code that allegedly steals credentials, API tokens…

Meta announced security upgrades to the hardware infrastructure protecting encrypted backups for WhatsApp and Messenger. The system stores recovery codes in tamper-resistant hardware security modules—specialized chips designed to resist…

Privacy & Surveillance · Meta

OpenAI launched Advanced Account Security, an opt-in protection tier for ChatGPT and Codex users facing elevated digital threats. The setting bundles passkeys or hardware security keys, disables email and SMS account recovery (common…

Privacy & Surveillance · OpenAI

Security researchers at Theori disclosed 'Copy Fail' (CVE-2026-31431), a Linux kernel vulnerability that has reportedly existed since 2017, allowing any unprivileged local user to gain root access via a simple 732-byte Python script. The…

The White House has told Anthropic it disagrees with the company's plan to give roughly 70 companies and organizations access to Mythos, the AI system Anthropic itself describes as powerful enough to enable dangerous cyberattacks…

Regulation & Policy · AI Safety & Risk · Anthropic

OpenAI released a cybersecurity policy document outlining how it believes AI should reshape digital defense. The 'Action Plan' proposes five pillars: making AI-powered security tools more widely accessible, coordinating government-industry…

Regulation & Policy · OpenAI

Apple patched a bug Wednesday that let law enforcement extract supposedly deleted messages from iPhones — including disappearing Signal messages. The flaw: when encrypted messaging apps displayed content in notifications, iOS cached that…

Privacy & Surveillance

A small group of unauthorized users has been accessing Mythos, Anthropic's most powerful—and most restricted—new AI model, since the day Anthropic announced its limited release, Bloomberg reported. Anthropic describes Mythos as capable of…

AI Safety & Risk · Anthropic

Vercel disclosed a security breach stemming from compromised third-party AI tool Context.ai. Attackers reportedly gained access to a Vercel employee's Google Workspace account, then accessed environment variables that weren't marked…

Meta is publishing its playbook for migrating to post-quantum cryptography, including a proposed framework of 'PQC Migration Levels' to help organizations manage the transition. The guidance references NIST timelines targeting 2030 for…

Meta

OpenAI announced a 'Trusted Access for Cyber' program, offering security firms and enterprises access to a specialized model called GPT-5.4-Cyber along with $10 million in API grants. The company says the initiative aims to strengthen…

OpenAI

A security researcher claims Fiverr has left customer files—including tax documents with Social Security numbers—publicly accessible and indexed by Google. The alleged exposure stems from Fiverr's use of public URLs rather than expiring…

Privacy & Surveillance

OpenAI announced GPT-5.4-Cyber, a specialized model available exclusively to vetted cybersecurity defenders through its Trusted Access for Cyber program. The company says it's expanding access to advanced AI capabilities for security…

OpenAI

Researchers released LogicDS, the first curated dataset of 86 real-world logical vulnerabilities (all with assigned CVEs), alongside a framework called LogicEval for testing how well automated repair tools—including LLMs—can fix them. The…

A buyer who acquired 30+ WordPress plugins through marketplace Flippa in late 2024 allegedly planted backdoors in all of them immediately after purchase. According to forensic analysis, the malicious code in one plugin—Countdown Timer…

Security firm AISLE claims it replicated much of the vulnerability analysis from Anthropic's recent Mythos announcement using small, inexpensive open-weights models. In their testing, all eight models—including one with just 3.6 billion…

Open-Weight Models · Anthropic

A security researcher found that Apple's two-VM limit for running macOS guests on Apple Silicon is enforced deep in the kernel, not in software that users can easily modify. However, they discovered undocumented boot arguments that can…

Little Snitch, a network monitoring and firewall tool long popular with Mac users for its ability to see which apps are phoning home, is now available for Linux. The Linux version provides per-application connection monitoring, traffic…

Privacy & Surveillance

VeraCrypt, the widely-used open-source disk encryption tool, says Microsoft terminated its developer account without warning or explanation—blocking all future Windows releases. The project claims it received only automated responses when…

Microsoft

Anthropic launched Project Glasswing, a coalition with AWS, Apple, Google, Microsoft, and eight other major tech companies to hunt software vulnerabilities in critical infrastructure. The initiative uses an unreleased model called Claude…

AI Safety & Risk · Anthropic · Google · Microsoft

Anthropic released a System Card for Claude Mythos Preview, a model it has decided not to make generally available due to capability concerns. The benchmarks are striking: 93.9% on SWE-bench Verified (vs. 80.8% for Claude Opus 4.6), 77.8%…

AI Safety & Risk · Anthropic

Researchers have proposed Human Delegation Provenance (HDP), a cryptographic protocol designed to verify that AI agent actions trace back to genuine human authorization. Published as an IETF Internet-Draft—an early step toward becoming an…

AI Agents

A former Azure Core engineer published a detailed account alleging that technical and leadership decisions eroded trust in Microsoft's cloud platform. The engineer claims that hosting a web service directly reachable from guest VMs on the…

AI & Jobs · Chips, Data Centers & Energy · Microsoft

Cloudflare released EmDash, an open-source content management system it calls a 'spiritual successor to WordPress,' built with AI coding agents over two months. The key pitch: solving WordPress's notorious plugin security problem. Instead…

AI Agents

Security researchers report that attackers hijacked axios—a software component so fundamental to the modern web that it is downloaded over 100 million times per week and is embedded in millions of websites, apps, and corporate tools. Think…

Anthropic accidentally published readable source code for Claude Code in an npm package—their second unintended exposure in a week after a model spec leak. Before the package was pulled, developers mirrored and analyzed the code…

AI Business & Money · Anthropic

Anthropic confirmed it is testing a new AI model called "Claude Mythos" after Fortune reported that draft materials about the system were left in an unprotected, publicly accessible data store on the company's website. An Anthropic…

AI Business & Money · Anthropic

OpenAI launched a Safety Bug Bounty program that pays researchers to find AI-specific vulnerabilities—not just traditional security holes, but risks like prompt injection, data exfiltration, and ways to abuse agentic AI systems. The…

AI Safety & Risk · AI Agents · OpenAI

A security researcher built a working Tesla Model 3 computer setup on their desk using salvaged parts from crashed vehicles, spending roughly $375-$475 total. The project—assembled from eBay-sourced components including the main computer…

LiteLLM, a popular open-source tool that lets developers route requests between different AI models (Claude, GPT, Gemini, etc.) through a single interface, was reportedly compromised on PyPI. Versions 1.82.7 and 1.82.8 allegedly contain…

The telltale sign of spam—sloppy design and broken English—is disappearing. AI coding tools now let non-technical scammers create polished, professional-looking fraudulent emails and websites. An Anthropic report found non-programmers…

Deepfakes & Misinformation · Anthropic

The European Space Agency ran a competition challenging 200+ teams to find hidden backdoors in AI forecasting models used for spacecraft telemetry. The concern: attackers could embed triggers in training data or model weights that cause…

Google, Amazon, Anthropic, Microsoft/GitHub, and OpenAI are collectively pledging $12.5 million to the Linux Foundation's Alpha-Omega Project, which funds open source security work. The investment aims to help maintainers defend against…

Google · Anthropic · OpenAI · Microsoft

Codex Security published a technical explanation of why its AI-powered security tool skips traditional Static Application Security Testing (SAST)—the scanning approach used by established tools like Snyk and Checkmarx. The company claims…

AI Business & Money · OpenAI

A technical deep dive explains how modern video game anti-cheat systems work at the kernel level—the highest privilege layer of Windows. These systems intercept low-level operating system callbacks and scan memory structures to detect…

Meta's security team says it has built a system that combines secure-by-default coding frameworks with generative AI to automate security fixes across its Android codebase. The approach reportedly lets AI propose, validate, and submit…

AI Adoption · Meta

Hacked data allegedly from the Department of Homeland Security's Office of Industry Partnership has surfaced online, listing 6,681 organizations that applied for DHS contracts. The leak includes contract details like award amounts, program…

Privacy & Surveillance

OpenAI published documentation on how ChatGPT defends against prompt injection—attacks where malicious instructions hidden in documents or websites try to hijack AI agents into taking unauthorized actions. The company says it constrains…

AI Agents · OpenAI

Researchers have proposed IH-Challenge, a training approach designed to make AI models better at following a hierarchy of instructions—prioritizing commands from system operators over potentially conflicting user prompts. The method claims…

AI Safety & Risk

Meta published technical details on how Messenger's Advanced Browsing Protection scans links for malware without compromising end-to-end encryption. The system uses private information retrieval—a cryptographic method that lets your device…

Privacy & Surveillance · Meta

OpenAI is acquiring Promptfoo, a startup that builds tools to help companies find and fix security vulnerabilities in AI systems before deployment. The deal signals OpenAI's push to address enterprise security concerns as more companies…

AI Business & Money · OpenAI

Researchers released OSS-CRS, a framework that makes DARPA's competition-winning AI cybersecurity tools usable outside their original cloud environment. The original systems from the AIxCC competition—where AI autonomously found and fixed…

Military & National Security

Wikipedia went into read-only mode on March 5, disabling editing across its wikis before restoring full functionality the following day. The official status page confirmed the outage but provided no details on the cause. Community…

A Hugging Face Space appeared claiming to offer free unlimited access to Google's Veo 3 video generation model. The listing provides no details about how it actually works or whether it delivers on the promise. Google's Veo 3 is currently…

Google

Anthropic's Claude Opus 4.6 discovered 22 vulnerabilities in Firefox over two weeks, including 14 rated high-severity—nearly 20% of all high-severity Firefox bugs remediated in 2025. The AI found its first critical memory flaw just 20…

Anthropic

OpenAI launched Codex Security in research preview, an AI agent designed to find and fix security vulnerabilities in codebases. The company claims it analyzes full project context to detect issues, validate whether they're real threats…

AI Agents · OpenAI

An unidentified hacker used Anthropic's Claude to orchestrate a series of attacks against multiple Mexican government agencies between December and January, stealing 150 gigabytes of sensitive data—including documents related to 195…

AI Safety & Risk · Anthropic

Security researchers report that Google API keys—long treated as non-secret identifiers for public services like Maps—can now silently authenticate to Gemini endpoints when the AI service is enabled on the same cloud project. Scanning…

Google

Google is embedding Gemini AI throughout Samsung's upcoming Galaxy S26, turning Android into what it calls an 'intelligent system.' New features include voice-activated task automation (ordering rides, food, groceries through partner…

AI Agents · Google

OpenAI released a threat intelligence report examining how bad actors combine AI models with websites and social media to conduct malicious operations. The report focuses on detection methods and defensive strategies rather than specific…

Deepfakes & Misinformation · OpenAI

A team of 37 researchers gave autonomous AI agents access to real tools—persistent memory, email accounts, Discord, file systems, and shell execution—in a live lab environment for two weeks and methodically tried to break them. The agents…

AI Agents · AI Safety & Risk

A security researcher reportedly discovered that a smart sleep mask broadcasts users' brainwave data to an unsecured server, with all devices sharing identical credentials. The alleged flaw means anyone who accesses the system could…

Privacy & Surveillance

OpenAI announced two new security features for ChatGPT: Lockdown Mode and Elevated Risk labels. The tools are designed to help organizations defend against prompt injection attacks—where malicious instructions hidden in documents or web…

AI Safety & Risk · OpenAI

UC Santa Barbara researchers released VulnLLM-R-7B, an open-source model built on Qwen2 and designed specifically for detecting security vulnerabilities in code. The 7-billion-parameter model targets automated code analysis tasks—scanning…

Open-Weight Models

A GitHub issue claims Microsoft's billing system can be bypassed using a combination of 'subagents' with an agent definition—though details remain vague and unverified. The disclosure quickly became messy: commenters report a fake…

AI Agents · Microsoft

OpenAI announced Trusted Access for Cyber, a framework that will give vetted security researchers and organizations expanded access to its most advanced AI capabilities for cybersecurity work. The program appears designed to let defenders…

AI Safety & Risk · OpenAI

Meta has released a method for passwordless login on devices without accessible screens, like VR headsets. The approach, now live on Meta Quest devices, lets users authenticate with passkeys stored on their phones without scanning QR…

Meta

A developer released NanoClaw, a stripped-down alternative to popular AI agent tools that uses Apple's native container technology to isolate each AI chat session. The project claims better security than existing tools by sandboxing agents…

AI Agents

A technical blog post details how a hobbyist defeated a 40-year-old hardware copy protection dongle—a physical device that software would query to verify legitimate ownership. Using modern reverse-engineering tools, the author bypassed the…

An online discussion highlighted how RF remotes, TV-Be-Gone devices, and tools like Flipper Zero can control others' electronics without authorization. The thread raised questions about whether consumer electronics should require device…

A developer released NanoClaw, a 500-line TypeScript alternative to Clawdbot that claims better security through Apple container isolation. Unlike OpenClaw, which runs agents with broad permissions in a single Node process, NanoClaw uses…

AI Agents

A Hacker News thread discussing noise complaints devolved into users sharing methods for remotely controlling neighbors' electronics without permission. Commenters described using TV-Be-Gone devices, smartphone IR ports, and RF remotes to…