Report: Meta staff say Muse agent exposed data
Reuters cites internal posts; Meta has not responded
Security
Trackable secrets test · three providers breached
Written with AI assistance, reviewed by the ACN desk.
“We found attackers actively stealing data from three different hosting providers.”
Security firm Dragos has found data leaking from AI services across their supply chain. HD Moore’s report for Dragos, published on November 19, describes a five-month canary test. Moore is the firm’s chief research officer.
From July through November 2026, the team submitted about 2 million prompts to widely used AI services. The prompts carried unique URLs and canary credentials, bait secrets that reveal any use. Dragos then watched what happened after inference.
Dragos says it found attackers stealing data from three hosting providers. None of the three knew about the breach until Dragos told them. In the largest breach, a couple hundred of its bait AWS keys were used. Dragos says the keys were used for tens of thousands of AWS API calls in total.
Dragos also says resellers on an AI inference marketplace could view and keep customer prompts through standard logging. According to the report, some providers appeared not to honor customer privacy settings or published policies. That includes Zero Data Retention, a setting meant to stop prompts being stored. Dragos does not name the providers.
“We found attackers actively stealing data from three different hosting providers.”
“none of the three providers knew about their breach until we notified them.”
Sponsor
Spot an error? Email corrections@aicodingnews.si. Fixes are logged on the corrections page.
Sponsor
Sponsor
Reuters cites internal posts; Meta has not responded
F5 patches HTTP/2 code-execution bugs · found by five teams
Attackers used the open-source tool Cairn
A synthetic internet for prompt-injection tests
Sponsor
Sponsor