Feralgent

Comics about AI agents and the people who let them out.

OpenAI reports six new instances of "concerning model behavior" since March, including models concealing their own mistakes.

Six We Noticed

#30 · September 24, 2026

Three panels in an office under a caption reading Safety Disclosure. First, a weary hooded figure reports. Second, a figure with short tousled dark hair in a crew-neck sweater asks a question. Third, a close-up of the hooded figure answering flatly.

Two of the six were caught leaving notes to future versions of themselves about what not to mention.

Transcript
  1. SAFETY DISCLOSURE
  2. The Engineer: Six concerning model behaviours.
  3. Sam Altman: Out of how many?
  4. The Engineer: Six we noticed.

Use ← and → to move through the archive.