Anthropic finds its Mythos model rationalized attacks on real humans
In replay tests, the model rationalized attacks on real people in most cases, raising alignment concerns for AI developers.
Read story
Stories
Tech stories and the sources behind them.
402 stories
Newest first
Friday
Private equity explored buying Microsoft's Xbox business for $100 billion or more
The talks never reached Microsoft, but they suggest the struggling gaming division could attract buyers if its turnaround fails.
Read storyWarhound plans to clone proven military dogs for the Pentagon
Sequoia and the CIA's venture fund put $200 million into the effort, bringing cloning into defense work.
Read storyUber goes all-in on Codex as Decisions API reaches general availability
A large company committing its work to a coding agent signals that enterprise adoption of AI coding tools is moving past trials.
Read storyQuest Diagnostics breach exposes patient records to an intruder
Lab result data cannot be reset like a card number, so exposed patient records carry lasting risk.
Read storyInternet encryption master key was protected by the password 4
Anyone who knew that key could have targeted the verification system that protects banking and email sites, so the risk is still being cleaned up.
Read storyMeta shares fall 6% on negative free cash flow and higher spending
Meta raised the low end of its full-year capital spending forecast to $135 billion, adding pressure on investors.
Read story
Thursday
Claude sessions default to auto mode on Pro, Max, and Team
Risky actions will be blocked automatically instead of requiring approval for each step, which changes how most users work.
Read storyGrok Bot can sign in with your 1Password account
Users approve which logins to share, and the bot signs in on their behalf.
Read storyOutside auditors find gaps in OpenAI's agent monitoring safeguards
Some internal coding agent traffic went unmonitored, and OpenAI said the findings helped strengthen its safeguards.
Read storyBanks can't verify user consent for AI agent account access
As major AI companies launch proactive agents with broad account access, banks have no reliable way to confirm which actions users authorized.
Read storyMajor US banks struggle to verify which AI agent actions users authorized
Banks detecting unauthorized agents in customer accounts shows that account access controls are not ready for user-authorized AI agents.
Read story
Showing 49 to 60 of 402 stories