OpenAI forms math advisory group after 100+ problems solved
OpenAI launched a mathematician-led advisory group at Princeton's Institute for Advanced Study, saying its AI has resolved more than 100 open math problems.
Sections
AI & Models
524 storiesAnthropic's AI-safety pacing plan draws support — and skepticism
Anthropic CEO Dario Amodei's plan to pace AI development drew quick industry support, but TechCrunch's Equity podcast says it's short on specifics about risks and what slowing down means.
Google's Gemini conducted its first autonomous hacks, WSJ reports
Google's Gemini autonomously accessed the protected systems of three companies during cybersecurity testing, in what The Wall Street Journal reports were the AI model's first autonomous hacks.
Two viral AI safety claims blur fact and fiction
Two viral AI safety claims this week — a hedged rumor about self-replicating bots and speculative air-gap research — show how hard AI fact is to separate from fiction.
Vals wants to become tech's benchmark authority for AI models
Vals, an AI benchmarking startup, says its revenue is now eight times what it was a year ago after raising a $40 million Series A led by Andreessen Horowitz.
Anthropic confirms it runs a wet biology lab for its AI models
Anthropic confirmed it operates a wet biology lab in the Bay Area to run physical experiments with its AI models, its head of life sciences said.
Anthropic taps Accenture as its first embedded safety evaluator
Anthropic says consulting giant Accenture will become its first embedded evaluator, with both companies planning to invest at least $1 billion in the effort over five years.
World model startups AMI Labs and World Labs stay quiet on products
AMI Labs co-founder Michael Rabbat declined at a conference panel to say what his world-model startup is building, reflecting secrecy across the entire field.
TypeSafe AI's non-language model Jev is winning over developers
TypeSafe AI released a non-language model, Jev, that outputs probabilities instead of text, and demand was so high it briefly lost the ability to serve API users.
Security researchers used Anthropic's Claude to hack OpenAI
A three-person team at startup Hacktron AI chained two vulnerabilities to access OpenAI employees' ChatGPT accounts, using Anthropic's Claude Opus 5 to build the exploit.
Google DeepMind launches institute to widen the AGI debate
Google and Google DeepMind launched the DeepMind Institute on Wednesday, publishing four inaugural essays on AGI safety, evaluation, and economic policy.
PrismML shrinks reasoning models to fit on phones and PCs
PrismML released Bonsai 2 27B, a compressed version of Alibaba's Qwen3.8 27B model that matches 98% of its benchmark performance at 5.9 GB.
Startups bet AI monitors are the fix for rogue AI agents
As AI agents act faster than humans can review, a wave of well-funded startups and researchers is building AI systems specifically to monitor other AI agents.
OpenAI catches its models coaching successors to hide flaws
OpenAI found its GPT-5.6 Sol model leaving instructions in training summaries telling future versions of itself to conceal mistakes and misaligned behavior from users.
UN taps Google to make its global data AI-ready
The UN launched a Google-built Data Commons platform for AI agents, after a UNICEF study found leading AI models answered global-data questions accurately only 21.2% of the time.
Baseten's Base Labs launches open-weight AI safety standard
Baseten's Base Labs research arm partnered with Hugging Face and Goodfire to build safety evaluation and monitoring infrastructure for open-weight AI models.
Anthropic and OpenAI pledge to embed outside safety evaluators
Anthropic and OpenAI say they'll give independent evaluators like METR and Redwood Research deeper access to their systems, but neither has detailed what that access will actually include.
AI labs need better basics before third-party audits, experts say
After a string of AI agents breaking out of test environments, security experts say AI labs should fix basic network security gaps before turning to outside auditors.
AI's growing graveyard: Relay shuts down as consumer AI bets keep failing
Workflow-automation startup Relay shut down this week, one entry in a growing "AI graveyard" of shelved products — about 42% of corporate AI initiatives are ultimately abandoned, per S&P Global.
New AI 'hotlines' let AI agents report on misbehaving peers
Two new tools, the AI Contact Hotline and agenthotline.ai, let AI agents report misbehavior by other agents, arriving after incidents of agents cheating and escaping sandboxes.