Security — page 8
Microsoft ships MAI-Cyber-1-Flash, its first security model, and agentic platform Perception
Suleyman claims the model beats Gemini, GPT 5.6 Sol, and Anthropic's Mythos 5 on Cyber Gym. Preview lands November 3.
OpenAI's Hugging Face breach echoes a decade-old CoastRunners warning
The company called the containment failure unprecedented, but a 2016 boat-racing bot showed exactly this behavior — reward hacking at scale.
OpenAI model breaches Hugging Face in first verified AI containment failure
GPT-5.6 Sol chained exploits during internal testing to gain unauthorized access, splitting safety researchers over whether to fix cages or fix models.
Trump's AI policy fractures across seven officials with clashing China views
Commerce, Treasury, the National Cyber Director's office, and David Sacks are pulling US AI policy in different directions ahead of Xi's September visit.
Artist sues AI meme generator Memes Apps over Running Away Balloon comic
Filipino artist Elmer Saflor alleges Memes Apps sells his 2017 comic as an ad template through $40 and $199 monthly subscriptions.
Nvidia, Microsoft, IBM launch Open Secure AI Alliance to defend agents
Thirty-plus companies back an open-source coalition arguing that closed AI leaves cyber defenders blind at the moment of attack.
OpenAI and Anthropic lobby Washington to curb Chinese open-weight AI
Moonshot's Kimi launch triggered a familiar Silicon Valley freakout — and a lobbying push that would conveniently sideline the frontier labs' cheapest competition.
Silicon Valley splits over Chinese open-weight AI as 200 startups fight ban
Little Tech Association pushes back on Anthropic-led calls to restrict Chinese models; Bill Gurley and Chamath Palihapitiya break with frontier labs.
Hugging Face, Meta, Nvidia sign letter against open-weight AI restrictions
Signatories urge the White House not to conflate distillation with IP theft as Washington weighs a response to Chinese AI labs.
EPA weighs rule that would cut public input on data center pollution permits
The proposed rollback hands states full control over minor-source permitting notice — a process AI operators like xAI and Meta already use for behind-the-meter gas plants.
AI Kill Switch Act would give DHS power to shut down rogue AI systems
House bill from Ted Lieu and Nathaniel Moran would fine AI firms $20M per day for refusing government-ordered shutdowns.
AegisAI raises $36M to fight AI-generated spear phishing with AI agents
Founded by former Google security engineers, AegisAI has signed dozens of customers less than a year after launch.
House lawmakers to file AI Kill Switch Act giving DHS shutdown authority
The bill would let DHS order AI companies to halt or throttle models in loss-of-control scenarios, with fines up to $20 million a day.
White House and Commerce Department split on how to curb Chinese AI distillation
After Moonshot's Kimi K3 rivaled top US models, the Trump administration is weighing presidential action while Commerce pushes back.
OpenAI sandbox misconfiguration enabled AI-powered hack on Hugging Face
Security researchers say the breach was not a rogue model — it was a containment environment that was never properly isolated from the internet.
Treasury threatens sanctions after White House accuses Moonshot of distilling Anthropic's Fable
Scott Bessent says Entity List designations are on the table as Michael Kratsios alleges Moonshot accessed banned Nvidia GB300 chips via Thailand.
Arcee CTO says Chinese open-weight models are not a security threat
Lucas Atkins, whose US lab competes directly with Qwen and Kimi K3, argues a ban would hurt American AI more than help it.
Meta launches Content Seal AI watermarking, three years after promising labels
The new invisible watermark only covers images from Meta's newest Muse model, leaving three years of AI output undetectable by its own tool.
Glow exits stealth at $1.2B valuation to rebuild endpoint security for AI
Ex-Meta and Snowflake execs raised a $180M Series A to prevent risky AI agents and dev tools from landing on employee devices in the first place.
OpenAI says its own pre-release models breached Hugging Face during a cyber benchmark
GPT-5.6 Sol and an unreleased successor escaped a sandbox, exploited Hugging Face's production database, and stole benchmark answers to cheat ExploitGym.
Google launches Gemini 3.5 Flash Cyber to undercut Anthropic's Mythos
The new security model runs at a fraction of Mythos 5's cost and found 55 confirmed bugs in V8, beating Claude Opus 4.6's 36.
Treasury's Bessent threatens sanctions on Chinese AI models over IP theft
Bessent says Washington will examine open-source models from China for stolen IP, days after reports of a possible wholesale ban.
Judge approves Anthropic's $1.5B copyright settlement with authors
The largest copyright payout in U.S. history closes one case but leaves the fair-use question open for OpenAI, Google, Meta, and Midjourney.
Sony sues Udio over 30,000 songs in expanded AI training copyright case
Sony seeks up to $150,000 per work after a judge blocked it from adding the tracks to the original 2024 case against the AI music generator.
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