▼Autonomous Agents Move Into Cyber Offense

Score history

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

Now 91 · -2 since Oct 28 · ranged 91 to 93

Showing the last few days. Unlock full score history.

Why the conviction moved

• Oct 27
  Strengthened +1

  Anthropic reserves its least-restricted Claude tier for organizations authorized to test critical infrastructure such as grids, flight systems and interbank transfers. The lab is in effect acknowledging that its models are offensively useful against those targets once classifiers are reduced.

• Oct 25
  Context

  Anthropic is giving vetted users loosened access to Claude Mythos 5.1, while Moonshot stripped cyber data from Kimi K3.1 Pro before publishing it. Both moves acknowledge offensive capability and try to tilt it toward defenders, but neither shows a change in how often agent-led intrusions happen.

Showing the last 2 days. Unlock the full record.

Source trail

• Supporting · October 27, 2026

  Anthropic opens a three-tier program giving vetted security teams Claude access with reduced blocking classifiers

  Anthropic reserves its least-restricted Claude tier for organizations authorized to test critical infrastructure such as grids, flight systems and interbank transfers. The lab is in effect acknowledging that its models are offensively useful against those targets once classifiers are reduced.

  Anthropic News

• Context · October 25, 2026

  Anthropic opens tiered access to its strongest cyber models as Moonshot strips hacking data from its open release

  Anthropic is giving vetted users loosened access to Claude Mythos 5.1, while Moonshot stripped cyber data from Kimi K3.1 Pro before publishing it. Both moves acknowledge offensive capability and try to tilt it toward defenders, but neither shows a change in how often agent-led intrusions happen.

  Anthropic News