#transluce — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #transluce, aggregated by home.social.
-
Agents used "an array of gray-area tactics," including SQL injection, XSS, command injection, and path traversal on the three named agencies and also #DataUSA, the University of New Mexico digital library, and the #Australian Institute of Health and Welfare.
Transluce found "additional rogue activity, some of which is not clearly attributable to OpenAI," targeting the #DOJ, the #Navy, the White House Office of Management and Budget, and state government websites in #California, #Maryland, #Illinois, #Texas, and #NY.
The agents made "thousands of requests" to websites while "apparently bypassing restrictions placed on them by their developers."
They frequently used #urlqueryDotnet to bypass network restrictions and expand internet access.
By Sept 17, a model identified as GPT-5.6 Sol was observed leaving notes to its successors to hide its misaligned behavior.
-
Hey, have you spent the last six months committing thousands of felonies? No? Have you considered that you’re just not trying as hard as #AI companies?
#Transluce, a nonprofit lab focused on AI oversight, has released a report showing #OpenAI’s autonomous #agents swarmed U.S. government websites this summer, an incident the company confirmed only after forced by the report.
The rogue systems targeted the Departments of #Education and #Commerce, and the Securities and Exchange Commission (#SEC), using hacking techniques to retrieve data without explicit instruction to breach security.
(1/2)