← @bgSponge

Countering misuse of AI: September 2026 / Anthropic

Summary

Anthropic's Threat Intelligence team disrupted malicious Claude operations across seven harm areas between late 2025 and 2026, identifying a shift where AI automates the cyber kill chain and inverts costs onto defenders.

  1. AI collapses the capability gap.

    The labor and tooling gap between well-resourced state-sponsored operations and individual operators has vanished, allowing hacktivists to function as advanced persistent threats.

    AI has collapsed the labor and tooling gap

  2. Attack economics favor the adversary.

    AI autonomy compresses the cost side of attacker ROI by handling reconnaissance and exploitation at machine speed, while leaving potential payoffs unchanged.

    AI autonomy compresses the cost side of attacker ROI calculations

  3. Exploit foundries automate vulnerability research.

    Threat actors establish autonomous workflows using Claude to conduct continuous, agentic vulnerability research and exploit design around the clock.

    effectively established automated exploit foundries with AI.

  4. Agent swarms maintain campaign memory.

    The GTG-10007 group used lead AI agents to decompose tasks and dispatch them to subagents, saving credentials and instructions across sessions.

    The operation maintained persistent campaign memory.

  5. Influence-as-a-service scales deception.

    Commercial actors use Claude to build entire operational apparatuses, including persona systems and target databases, to mass-produce political content.

    used Claude to mass-produce and rewrite political content.

Read the original ↗Save to your Sponge

Summarized with Sponge. Quotes are the author’s own words.

More from @bg

All readings →