UK Pays $400K for AI That Calls 13-Year-Olds Adults
The UK spent $400K on face-scanning AI it knows misidentifies 13-year-old girls as adults, and still plans to deploy it on asylum seekers by 2027.
AI-compliance: SEC AI-washing, FTC Operation AI Comply, FDA AI/ML SaMD, state AI laws.
The UK spent $400K on face-scanning AI it knows misidentifies 13-year-old girls as adults, and still plans to deploy it on asylum seekers by 2027.
The NO FAKES Act cleared committee: $750K per deepfake, heir rights for at least 70 years. Civil liberties groups warn it will gut satire and parody online.
Lutnick threatened Anthropic over a Fable 5 jailbreak claim with criminal and civil penalties, setting a bar that would freeze all frontier AI.
The DOJ's first domain seizure under the TAKE IT DOWN Act took down CFAKE and SOCFAKE, with a French suspect already arrested in Nice.
A multistate AG coalition served OpenAI a sweeping subpoena four days after its S-1 filing, covering kids, health data, and model sycophancy.
A disputed jailbreak report led Commerce to put Fable 5 and Mythos 5 under export controls, cutting off Anthropic's own foreign-born employees.
Amazon's biggest backer tipped off the feds to a Fable 5 jailbreak, and Dario Amodei refused the White House's fix-or-pull ultimatum.
OCC and Fed examiners are now asking every bank about AI kill switches, vendor subcontractors, and data access controls. The formal RFI is next.
DOJ and HSI seized two deepfake porn sites serving 4 million monthly views, and a French admin's $48K Ethereum stash led to an arrest in Nice.
A 93% AI confidence score isn't a match probability. Florida cops knew it, filed the warrant anyway, and jailed Robert Dillon for two months.
Munich's Regional Court ruled Google owns its AI Overviews speech, meaning it must pay when that speech defames publishers with invented facts.
Meta's AI-powered account recovery tool didn't check email ownership, and attackers used that gap to steal 20,225 Instagram accounts.
A Nashville teen sues Omnilert over a $1M AI gun detector that didn't fire during a school shooting the company claimed its tech could have prevented.
Anthropic deployed six engineers to the NSA to run a model the DoD officially banned, a contradiction both sides are declining to explain.
Trump's June 2 EO hands feds up to 30 days pre-release access to frontier AI models. OpenAI said yes first, and the voluntary label won't last.
A 269-page bipartisan draft would freeze state AI laws for 3 years, but the $500M audit threshold and sunset clause are where operators look.
XBOW found an API key in Moderna's source code, authenticated itself, and crashed the entire dev stack in hours. Mythos just made this repeatable.
ETSI TS 104 033 sets GPU isolation and Model BoM rules for AI data centers. EU regulators routinely turn ETSI specs into enforceable law.
UK regulators gave Google nine months to give publishers AI opt-out controls, and banned Google from downranking any publisher who opts out.
A class action seeks $5M over Ring's Familiar Faces: bystanders scanned without consent can't opt out without tracking down each camera owner.