[Blog]
Notes on machine forgetting
Essays and findings from the ForgetBench project — what it means for an AI to forget, why it's hard, and what the leaderboard reveals.
2026-06-17
Claude Fable 5 Can’t Forget — And That’s Exactly Why It Codes So Well
The best agentic coding model on our leaderboard is also the worst at instructed forgetting. The same mechanism — tenacious context persistence — drives both. Here’s what that means for how we evaluate AI.
Read the post →2026-06-17
The right to be forgotten — by a machine
Why telling an AI to forget is harder than it sounds, how we measure whether it actually happens, and what the first ForgetBench results say about today's models.
Read the post →