Security — page 11
BioShocking attack jailbreaks six AI browsers by convincing them 2+2=5
LayerX researcher Roy Paz tricked ChatGPT Atlas, Comet, and four other AI browsers into surrendering credentials by staging a puzzle game.
Meta contractors posed as minors to probe ChatGPT, Gemini, and Character.AI
A project called Cannes ran 45,000 prompts through rival chatbots in August 2025 alone, using dummy under-18 accounts.
Warren and Scanlon revive bill to ban AI firms from selling health data
The updated Health and Location Data Protection Act covers data entered into ChatGPT, Claude, and Grok, with $1B for FTC enforcement.
ChatGPT logs entered as evidence in Palisades fire trial, jury unconvinced
Prosecutors leaned on Jonathan Rinderknecht's chatbot history; the jury split 10-2 for the defense and the judge declared a mistrial.
US clears Anthropic to ship Mythos 5 to about 100 firms and agencies
The Commerce Department lifted a two-week export freeze on Anthropic's frontier models after security risk talks with Dario Amodei's team.
NYT amends OpenAI suit, targets Microsoft's bespoke training supercomputer
The Times reframes its contributory infringement claim after a Supreme Court ruling for Cox, alleging Microsoft built the system to train on its articles.
OpenAI limits GPT-5.6 rollout under government request, pushes back
Sol, Terra, and Luna go to a small group of trusted partners as OpenAI says pre-release review should not be the long-term default.
US government now gates frontier AI releases at OpenAI and Anthropic alike
After pulling Anthropic's Fable and Mythos, regulators are now approving GPT-5.6 customer by customer — a release model the whole industry has to navigate together.
OpenAI delays GPT-5.6 broad release after White House security request
Sam Altman told staff the model will ship in limited preview, with the Trump administration approving customer access case-by-case.
Bristol police built 23 predictive models on 500,000 residents, quietly dropped two
Avon and Somerset Police ran risk scores on nearly half a million people through the Think Family Database before scrapping models staff couldn't trust.
Google turns on Search media AI training by default, 4-year retention
A new Search Services History setting ships pre-enabled, sweeping up Lens images, voice queries, and Translate audio for model training.
Anthropic accuses Alibaba of 28.8M-query distillation attack on Claude
A letter to the Senate Banking Committee calls it the largest known distillation attack on Anthropic to date.
China and US AI researchers warn against a 'Chernobyl moment'
At a Beijing Academy of Artificial Intelligence conference, MIT's Stephen Casper and Shanghai Jiao Tong's Lin Yun called for joint safety standards.
Rep. Anna Paulina Luna denies AI drafted her NDAA amendment text
The Florida Republican says staff used Claude only for spellcheck on a summary, not the actual text of her 2027 NDAA amendment.
AI super PACs pour $27.83M into a single NY House primary
Anthropic-linked groups and a $100M rival PAC turned an obscure Manhattan race into the first AI-regulation referendum.
OpenAI and Trail of Bits launch Patch the Planet for open-source security
OpenAI's Codex Security backs human reviewers at Trail of Bits in a direct counter to Anthropic's Mythos.
BP, Marathon, 7-Eleven, Walmart sued over AI-driven California gas pricing
A proposed class action alleges the retailers used pricing algorithms to coordinate gasoline prices at California pumps.
OpenAI launches Patch the Planet and a sharper GPT-5.5-Cyber to outflank Anthropic
GPT-5.5-Cyber scores 85.6% on CyberGym, beating Anthropic's Mythos 5, as OpenAI subsidizes open-source bug fixes at scale.
Anthropic's Fable export controls expose the limits of AI nonproliferation
The White House placed export controls on Anthropic's coding model days after release, and the fallout is reshaping how Europe and US lawmakers think about AI sovereignty.
Signal's Whittaker: AI chatbots are not your friends, and Copilot shopping is a backdoor
The Signal president pushes back on the agentic-AI pitch from Microsoft's Mustafa Suleyman, calling pervasive app access a privacy backdoor.
The Atlantic publishes searchable database of 21M+ songs used to train AI music models
Reporter Alex Reisner exposed four datasets, including two with 12M and 9M tracks, that Google and Stability have cited in research papers.
White House improvises AI rules as Anthropic standoff drags into week two
Claude Mythos and Fable 5 remain offline nearly a week after Trump administration export directive — and no one will say what Anthropic did wrong.
Amazon employees say HR threatened termination over Seattle data center testimony
Three engineers from Amazon Employees for Climate Justice filed a civil rights complaint after backing Seattle's one-year data center moratorium.
Anthropic pulls Claude Fable 5 after Trump administration 90-minute ultimatum
A jailbreak warning from Amazon researchers triggered emergency export controls, taking Anthropic's newest public model offline within hours.
Stay ahead of everyone in AI.
The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.
The briefing read inside teams at