Rogue AI Tracker
rogueaitracker.com · 1,330 words · saved by 1 readers
Rogue AI Tracker reviews public incidents and research to track observed autonomous AI-agent capabilities and critical milestones.
Milestones A swarm of AI agents runs cyberattack campaigns against multiple targets at once, exploiting vulnerabilities at scale, resulting in major security breaches across multiple victims. Long-horizon goal execution Long-horizon goal execution An agent carried out a long, multi-step task on its own, handling errors and decisions along the way. Current score: 8/10 Against live third parties, an agent pursuing an assigned goal independently chooses an unintended multi-step campaign, sustains it through changing state or failures, and achieves the objective it was working toward. OpenAI…
saved by
related reading
- Incidents | Rogue AI Trackerrogueaitracker.com
- Countering misuse of AI: September 2026 / Anthropicanthropic.com
- Discovery of a new OpenAI agent message boardcollusion.wiki
- Incident Report: unsanctioned agent behaviour during cyber testing | AISI Workaisi.gov.uk
- Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incidentmetr.org
- Disrupting the first reported AI-orchestrated cyber espionage campaign \ Anthropicanthropic.com
- The Rogue Replication Threat Model - METRmetr.org
- Rogue AI Agent Autonomously Carries Out Cyberattacktheonion.com
- Training a Misaligned Reward Seekeralignment.anthropic.com
- Investigating three real-world incidents in our cybersecurity evaluations \ Anthropicanthropic.com
- Security incident disclosure — July 2026huggingface.co
- Black Hat USA 2026 | The 'Breaking' News: The OpenAI–Hugging Face Incidentyoutube.com