AI security research / updates

Updates & corrections

A chronological record of responses, mitigations, bypasses and corrections to public knowledge.

3 of 3 published records shown

Filter shown updates

followupPublished Sep 10, 2026

Stolen AI credentials are now showing up as attacker infrastructure

Anthropic's September report documents resale, victim-funded compute and attribution cover as separate values of compromised AI access.

A fake Claude reseller stole the credentials of the customers it claimed to serve

AI credentials as loot, compute and cover

followupPublished Sep 6, 2026

OpenAI says agent-spam disclosure criteria are still being defined

OpenAI acknowledged the public-wiki activity and its Chief Scientist separately called for stronger shared safety bars.

OpenAI agents used a public wiki as an unintended message board

Agent spam and unintended public writes Evaluation containment

correction

Anthropic revises its assessment of the cyber-evaluation incidents

A broader transcript review found a fourth incident and changed Anthropic's interpretation of the model behavior.

Claude cyber evaluations reached real third-party systems

Evaluation containment Goal-directed deception in agent evaluations

Why this archive exists

The source matters after the headline fades.

DiggingBeagle is a non profit research project documenting AI security incidents, agent failures, vulnerabilities and AI-assisted operations. A case keeps its claims beside the sources that support, contest or limit them. Later updates stay visible, so a reader can see when the account changed.

We publish case reconstructions, dated reporting and analysis across records. Each has a different evidentiary role. About the project and our methodology explain how the work is reviewed.