▲The Agent Skill Supply Chain Becomes an Attack Surface

Score history

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

Now 85 · +2 since Oct 11 · ranged 83 to 85

Showing the last few days. Unlock full score history.

Why the conviction moved

• Oct 12
  Strengthened +2

  A new arXiv study catalogues prompt-injection routes through both direct user input and indirect channels, in single-agent and multi-agent setups. Indirect channels such as tools, retrieved content and other agents become places where untrusted instructions get in, which supports the push for zero-trust verification across the agent stack.

• Oct 10
  Strengthened +4

  Microsoft Execution Containers, enforced sandboxes for AI agents in Windows 11, reached general availability on October 7. That puts zero-trust isolation of agent tool execution into a mainstream operating system, the platform-level control this thesis predicts.

Showing the last 2 days. Unlock the full record.

Source trail

• Supporting · October 12, 2026

  New research catalogues prompt-injection routes through agent systems as Nadella says models should be treated as compromised

  A new arXiv study catalogues prompt-injection routes through both direct user input and indirect channels, in single-agent and multi-agent setups. Indirect channels such as tools, retrieved content and other agents become places where untrusted instructions get in, which supports the push for zero-trust verification across the agent stack.

  arXiv cs.CR

• Supporting · October 10, 2026

  Windows 11 ships enforced sandboxes for AI agents as an open-source tool hands agents a decompiler

  Microsoft Execution Containers, enforced sandboxes for AI agents in Windows 11, reached general availability on October 7. That puts zero-trust isolation of agent tool execution into a mainstream operating system, the platform-level control this thesis predicts.

  AI ML Big Data