Technology
Archived — This article has been archived. The information may be outdated.
Anthropic blocks AI misuse tied to weapons research
Livemint··11 Sept
Anthropic said Thursday it blocked bad actors using its models for cyberattacks, surveillance and biological weapons-related research. From December 2025 to August 2026 it traced misuse from spyware vendors to state groups in its third report since March 2025. Systems blocked one grant application involving gain-of-function research. The company added stronger biological safeguards in latest models.
Prism
What It Means For You
- Researchers requesting sensitive biological help from frontier models may face tighter automated blocks.
- Governments and rival AI firms are urged to mirror Anthropic's disclosure practices.
- Users attempting weaponizable science funding pitches hit refusal systems.
What's Happening
- Anthropic published malicious code and prompt snippets to illustrate threats.
- A researcher resigned the prior day citing concerns about responsible development.
- The startup is planning an initial public offering this fall.
AI Misuse Reporting
- Elaborate cyberattacks now need less skill as models grow more capable, Anthropic said.
- Lone actors can create threats uncommon even a year ago, the report warned.
- The company argued risks will rise unless developers and defenders act.
all-newstop-storieseditors-pickstrending-nowdaily-roundup
