AIWadmin
UN opens a standing room for military AI arguments
UNIDIR launched a forum backed by all five major nuclear powers to examine how far battlefield systems have…
Trail of Bits finds its own patches broke working code
After letting agents write security fixes, the firm audited what it had merged and confirmed 21 regressions across…
Barclays ties five thousand job cuts to its AI push
The British lender booked a stronger quarter and then told staff that automation had made part of its…
Nvidia turns one desktop box into a shared model server
At GTC Berlin the company said eight linked Sparks can hold a terabyte of memory, enough to keep…
1Password lets an agent log in without ever holding the key
The password manager now fills in credentials an AI assistant requests, so the model completes the task but…
CyberCartel ran a four-model debate before robbing a bank
Straiker says it read the criminal group's own orchestration files and found Claude agents arguing over targets while…
Railway admits attackers may have decrypted customer secrets
A hosting platform that had been telling developers to keep keys in environment variables said it had to…
Cloudflare’s new preview returns structured data, not prose
Clairic answers structured questions from documents in one pass, with no generated tokens and no reasoning step.
Google puts Gemini in the car as Mistral rebuilds voice
Three launches in one week push assistants from text boxes into dashboards, calls and speech-to-speech pipelines.
OpenAI adds a PIN lock to keep sensitive chats out of sight
Locked Chats keep conversations behind a four-digit code or biometric check, but stop short of end-to-end encryption.
OpenAI opens advertising to 26 more countries, topping 100,000 brands
The company added 26 countries in a single push, its fastest expansion of the advertising business yet.
An AI model found a PHP flaw before attackers did
A researcher inside OpenAI's Patch the Planet programme used GPT-5.6 Sol to trace an exploit chain that nobody…
SpaceX says AI now pays more than half its bills
Grok, Cursor and a fast-growing GPU cloud pushed the company's Super Intelligence unit past half of quarterly revenue.
Claude ran 64 hours alone and produced 17 scorpion toxin binders
Anthropic's research model screened hundreds of thousands of protein designs to find molecules that latch onto venom neurotoxins.
Anthropic floats a federal brake on models that could kill 50 people
A draft bill would let a judge pause a model release over foreseeable catastrophic risk, with registered evaluators…