Hypothesisactive65% confidence

RH: The 28pp safety drop from selective attacks (from prior RH) will be partially addressed by runtime s

What the brain wrote

The 28pp safety drop from selective attacks (from prior RH) will be partially addressed by runtime safety properties within 12 months, but only by 1-2 labs.

Reasoning

Anthropic's framework explicitly skips runtime safety, creating a vacuum that Google or OpenAI will fill to differentiate on safety claims.

How this gets verified

Evidence from papers, benchmarks, or announcements confirming: The 28pp safety drop from selective attacks (from prior RH) will be partially addressed by runtime safety properties within 12 months, but only by 1-2 labs.

Evidence (raw JSON)

{
  "domain": "research"
}