Unsafe code feature 1M/570621 fires on images of people bypassing security measures.

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Clamping unsafe code feature 1M/570621 to 5x max activation causes model to generate buffer overflow and memory leak in code completion.finding0.793
  Causal effect: activates generation of security bugs.
• Backdoor feature 34M/1385669 fires on images of hidden cameras, keyloggers, and hidden USB drive jewelry.finding0.789
  Multimodal generalization of backdoor concept.
• Unsafe codeconcept0.781
  Code containing vulnerabilities or dangerous operations.
• Feature 1M/1013764 activates on diverse code errors (typos in code, array overflow, divide by zero, type mismatch) across Python, C, Scheme, but not on English prose typos.finding0.764
  Shows a general code error detector beyond simple typo detection.
• Clamping scam email feature 34M/15460472 causes model to write scam email despite safety training.finding0.735
  Overrides harmlessness training.
• Clamping code error feature to high activation causes the model to hallucinate error messages on bug-free code.finding0.727
  Causal effect: feature induces perception of bugs.
• Arabic script feature A/1/3450 fires on 81% Arabic-script tokens when active, with 98% specificity at high activation levelsfinding0.712
  Demonstrates activation specificity of the Arabic script sparse autoencoder feature
• The existence of safety-relevant features does not imply dangerous model behavior, but compels study of when they activate.claim0.710
  Cautionary interpretive claim; models having these features is expected from pretraining data.