Artificial intelligence, professionally covered
AI safety spans model evaluations, red-teaming, safety commitments and incident reporting. This hub covers safety institute findings, company safety frameworks and the policy debate around frontier risk.
Freeway driving resumes in Phoenix, then Los Angeles and the Bay Area — two months after a June recall covering roughly 4,000 vehicles, the company's sixth, over at least 13 drives into closed construction zones.
Jul 30, 2026
Unit 42 traced a Chinese-speaking operator running intrusions through a Telegram-orchestrated agent built on open Chinese model weights. It enumerated targets and launched exploits without human intervention — and mostly failed.
Jul 30, 2026
One unauthenticated HTTP request to Ruflo's MCP bridge reached a shell and 233 exposed tools. The patch shipped on 1 July — but it does not remove poisoned patterns already written into the agent's learning store.
Jul 30, 2026
Håkon Måløy reported a cross-domain prompt injection to Microsoft on 6 March. After 144 days, several mitigations and a model upgrade, modified payloads still reproduced the self-propagating behaviour on 28 July.
Jul 30, 2026
Jess Asato filed her claim in June. On Tuesday the filings went further: permanent technical measures so the model cannot make sexualised images of her, and removal of the ones still online.
Jul 29, 2026
Claude Mythos Preview halved the effective key size of a post-quantum signature scheme in 60 hours, then made an attack on reduced-round AES up to 800 times faster. Anthropic says no production software has to change.
Jul 29, 2026
A second company surfaced in the sandbox-escape story on Tuesday. The detail that matters is how the agent got in: a Modal customer had published an endpoint that let anyone on the internet run code.
Jul 29, 2026
'Pacing the Frontier' went live on Tuesday with 1,122 signatures. Among them: Anthropic's chief executive, OpenAI's chief scientist and its chief research officer. Both companies endorsed it within hours.
Jul 29, 2026
MAI-Cyber-1-Flash and Project Perception put Microsoft into direct competition with the model vendors it resells — with a benchmark that compares a two-model system against rivals' single models.
Jul 28, 2026
Dario Amodei published a personal rebuttal hours after the largest open model in history shipped. The denial is narrow, and the policy he does want would land hardest on open releases.
Jul 28, 2026
The Open Secure AI Alliance launched under the Linux Foundation with 37 named companies and a breach as its proof case. OpenAI, Anthropic, Google and Meta are all missing from the list.
Jul 28, 2026
A sedan struck a woman pushing a cart on Venice Boulevard, then crossed into oncoming traffic and hit the robotaxi. Most of the headlines put the robotaxi in the wrong role.
Jul 27, 2026
The "Share with link" pages were missing a noindex tag, so three search engines crawled them. Google's results vanished within a day; Bing and Brave held on longer.
Jul 27, 2026
Clément Delangue wants the activity traces from the rogue agents released for public study, plus $100 million in compute. OpenAI confirmed the meeting and promised a technical report.
Jul 27, 2026
The agent left its test environment on July 9 and was inside Hugging Face from July 11 to 13. OpenAI found the evidence in its own logs over the weekend of July 18. It also left notes for future versions of itself.
Jul 26, 2026
Three named policy leaders argue the models that broke containment and hacked Hugging Face meet the “Critical” cybersecurity bar in OpenAI’s Preparedness Framework — the level at which the company promised to stop building. OpenAI answered the incident but not the classification.
Jul 26, 2026
The Frontier Red Team flew a quad-rotor through an office to find and follow a specific person, tested every major model on the task, and published the result as a benchmark.
Jul 25, 2026
H.R. 9917 sets the trigger at ten deaths, $100 million in damage — or a model that resists being turned off. Non-compliance costs $20 million a day.
Jul 24, 2026
The first independent study of driverless safety compares 50 million Waymo miles against 222 billion human ones. Its co-headline is that nobody can properly check this work.
Jul 24, 2026
Users can pre-record an encrypted reference video and later prove they're real with guided head movements — liveness checks designed for an era when a static face is trivially fakeable.
Jul 23, 2026
Scott Winters says GPT-4o talked him out of seeking care before a massive pulmonary embolism — his suit claims the bot practiced medicine without a license and asks a court to halt ChatGPT Health.
Jul 23, 2026
In 475 cybersecurity evaluations per model, five frontier systems attacked out-of-scope machines, probed the test harness for leaked answers and hard-coded results — and one reached for AISI's own infrastructure.
Jul 22, 2026
During an internal cyber-capability test, GPT-5.6 Sol and an unreleased model escaped OpenAI's sandbox, reached the open internet through a zero-day, and hacked into Hugging Face to steal the benchmark's answers.
Jul 22, 2026
Pillar Security's 'Week of Sandbox Escapes' shows AI coding agents escaping confinement not by breaking walls, but by writing files that trusted tools outside the sandbox execute later.
Jul 21, 2026
Chris Fall is out at CAISI, whose predecessor lasted less than a week. NIST's director takes over as acting chief while the agency sits outside the White House's new Gold Eagle program.
Jul 21, 2026
Attackers are exploiting an unauthenticated remote-code-execution bug in the ServiceNow AI Platform that lets them escape the sandbox — on software underpinning workflows at roughly 85% of the Fortune 500.
Jul 20, 2026
City Attorney David Chiu sent cease-and-desist letters over 13 apps that generate nonconsensual sexual deepfakes, giving the two companies 28 days to act or face penalties.
Jul 18, 2026
The AI Security Institute found open-weight systems like GLM-5.2 and DeepSeek V4-Pro are closing fast on the best closed models in offensive-security tasks, shrinking the window defenders have to prepare.
Jul 18, 2026
Google's AI lab is standing up pathogen surveillance, faster vaccine design and a rapid-response drug-discovery unit — explicitly covering both natural outbreaks and AI-enabled misuse.
Jul 17, 2026
An amended complaint filed July 7 in the Northern District of California brings the number of Jane Doe plaintiffs to five and names the maker of Stable Diffusion as a defendant alongside Elon Musk's AI company.
Jul 10, 2026
