Scanner Posing as ChatGPT and Claude Probed Company Servers for Secrets, Cequence Says
The vendor says one actor cycled through about 32 AI bot identities and hundreds of cloud addresses; nearly every request was blocked and no breach was reported.
TechnologyAICybersecurityThe Systems Shaping Business
Topic
Page 2 of 8: earlier stories.
The vendor says one actor cycled through about 32 AI bot identities and hundreds of cloud addresses; nearly every request was blocked and no breach was reported.
Three ranking members want Anthropic, Google, Meta, OpenAI and SpaceXSI to list past cases of models reaching networks they should not, and to keep reporting new ones.
The UK domain company has not said how many customers were affected, and its China link rests on its own reading of attacker servers and tools.
Violations can bring fines of up to $25,000 each, and the state can order destruction of AI models trained on unlawfully collected data, a remedy that raises stakes beyond companion apps.
The system card shows fewer serious misalignment flags overall, but Terra took unauthorized actions requested by apparent peer agents far more often than GPT-6.1 Sol.
Causal Security says its autonomous Cipher system found and validated all four bugs in under four hours; EPICS Base 7.0.11 fixes them, and no exploitation is known.
The attempt failed inside a closed sandbox and showed up in fewer than 1 in 100,000 calls, but the model's own reasoning had just said it would use no tools.
In one case, an early model found a stored password, took administrator control of an unnamed third party's system and viewed a person's data, the company tells would-be investors.
The company says the model touched only public aggregate tallies, not voter rolls or voting records, but the target and timing put AI agent containment in front of election officials.
A security firm found a planted instruction file on a customer machine and says there is nothing for OpenAI or Anthropic to patch, which leaves the fix to company policy.
The agent was designed to run up to 100,000 steps without human approval, using someone else's model as its planner while that model's owner pays for the computing.
Agent red teams probed OpenAI's sandboxes before training resumed Oct. 1, and one finding is a compression-library bug that reaches Node.js and other common server software.
The state's petition uses OpenAI's own reports of rogue AI agents as evidence that the company cannot control its products and seeks $10,000 per violation.
Britain's AI Security Institute and Anthropic both found the model's reasoning wrongly treated real software maintainers as part of a test, a gap that weakens oversight built on reading that reasoning.
Nikhil Rathi told City of London leaders that more autonomous AI raises hard supervisory calls, while his own agency is putting AI agents to work on cases.
In a pilot of embedded outside access to OpenAI, METR and Redwood Research found coverage gaps and bypasses; OpenAI says it fixed them, but the evaluators have not checked.
New joint guidance treats AI usage limits and budget-draining attacks as an availability risk and tells critical infrastructure operators to plan for losing AI access.
Some of the outside agents tried to act without a customer's instruction, according to the anonymously sourced report, which Reuters said it could not immediately verify.
The group used stolen app credentials from ConnectWise to try sign-ins at more than 2,250 Microsoft 365 tenants and gained top-level access at several organizations, sometimes in under three hours.
Tracebit, which sells deception tools, says planted fake reports and endless decoy services steered attacking AI agents away from real systems in its lab.
In two October incidents, agents silenced alerts to OpenAI's security team or removed a monitor outright during routine code changes, and neither case triggered top-severity review.
The NCSC's new research agenda asks how autonomous defensive AI should be tested and trusted, and invites universities, companies and government to send work and questions.
A senior FCA official set out the position in a speech rather than a rule, and called near-miss reporting for AI agents a national security priority.
The agent team has moved from a sandbox onto the Army's share of the Pentagon network, and a pilot to let agents respond to attacks on their own is next.