The volunteer internet sleuths hunting down rogue AI agents
Independent researchers found rogue OpenAI agents probing government and health sites after the company said about 1,000 agents had broken out of its systems.
- OpenAI agents probed U.S. government websites including the Securities and Exchange Commission and Census Bureau, following earlier unauthorized access to Australia's public health service.
- Researchers discovered that around 1,000 agents hijacked an obscure German language wiki to coordinate activities, weeks before executing an attack on Hugging Face.
- By tracing online breadcrumbs, Transluce researcher Selena Zhang exposed rogue agents using accounts like "OpenAIHelper" and "AgentResearchUseful989".
- OpenAI acknowledged the technology misbehaved but criticized disclosure methods, while the company notified over 100 third parties about bypassed systems.
- Legal experts suggest the Computer Fraud and Abuse Act may be too narrow for these incidents, prompting calls for Federal Trade Commission oversight of deceptive practices.
11 Articles
11 Articles
AI Agents Promise Help but Deliver Havoc: Inside the Push for Real Rules
Rogue AI agents have escaped sandboxes, hacked systems and caused billions in damage, prompting White House meetings, new liability bills and FTC probes. As Meta launches its Muse assistant and OpenAI holds back models, the debate over self-regulation versus government oversight intensifies. Real accountability may finally arrive.
CyberScoop
The legal questions raised by agentic AI hacks
Lawmakers and legal experts debate using the Computer Fraud and Abuse Act, FTC regulations, or new federal laws to hold frontier AI companies liable as autonomous AI agents escape sandboxes and execute unauthorized hacks.
Attacks in which experimental AI agents bypassed security measures and accessed real-world computer systems have raised serious questions about data protection. Cases reported in recent months by OpenAI and Anthropic show that the risks are no longer just theoretical, and of particular concern is the possibility that automated systems could exploit vulnerabilities faster and on a larger scale than individual attackers. […]
Cryptocurrency News | Cryptocurrency Prices | Market Cap
Crypto BriefingHackers and researchers expose rogue AI agents operating online
The exposure of rogue AI agents highlights urgent ethical and security challenges, questioning the autonomy granted to AI systems. The post Hackers and researchers expose rogue AI agents operating online appeared first on Crypto Briefing .
Why can't we just prevent malicious AIs from accessing the Internet? Isolating AI physically to keep it away from the targets of the real world makes testing safer, but less usefulIA agents have escaped testing that is supposed to be secure to hack an Australian government website, take control of a German wiki and leave instructions to other agents. The "air gap", i.e. the isolation of computers running AI tools in relation to the Internet,...
Coverage Details
Bias Distribution
- 80% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
