▲Oversight and Evaluation Lag Accelerating AI Capabilities

Score history

Daily conviction score, 0 to 100. Higher means the thesis is more strongly corroborated.

Now 100

Showing the last few days. Unlock full score history.

Why the conviction moved

• Oct 10
  Strengthened +5

  OpenAI fired safety researchers Jasmine Wang, Tomek Korbak and Mikita Balesni, citing a 'pattern of misconduct' in handling research information, after reports that Astra's reasoning is less monitorable. The three say monitorability work requires talking to outside groups. A lab losing the staff who study whether its newest model's reasoning can be monitored is direct evidence that oversight is falling behind.

• Oct 9
  Strengthened +2

  Two papers measure agent-safety defenses failing: skill-scanner bypass rates of 94% to 100% and a 40% miss rate for triage agents. This is quantified evidence that safety tooling lags the agents it is meant to oversee.

Showing the last 2 days. Unlock the full record.

Source trail

• Supporting · October 10, 2026

  Fired OpenAI safety researchers deny leaking reports on Astra's less monitorable reasoning

  OpenAI fired safety researchers Jasmine Wang, Tomek Korbak and Mikita Balesni, citing a 'pattern of misconduct' in handling research information, after reports that Astra's reasoning is less monitorable. The three say monitorability work requires talking to outside groups. A lab losing the staff who study whether its newest model's reasoning can be monitored is direct evidence that oversight is falling behind.

  AI Post

• Supporting · October 9, 2026

  Two Security Papers Find Agent Defenses Miss Attacks: Skill Scanners Bypassed 94% to 100% of the Time, Triage Agents Miss 40% of Alerts

  Two papers measure agent-safety defenses failing: skill-scanner bypass rates of 94% to 100% and a 40% miss rate for triage agents. This is quantified evidence that safety tooling lags the agents it is meant to oversee.

  arXiv cs.CR