Development

Imperceptible audio hijacks thirteen speech models

development2026-05-20

Claim

Researchers demonstrated imperceptible audio attacks hijacking thirteen audio language models at 79 to 96% success, while Anthropic reversed course and told its security partners it fully supports publishing their findings.

Reported metric: 79-96% success

Reporting date: 2026-05-20 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to May 20, 2026 (theinnermostloop.substack.com)

Knowledge graphopen in browser →

Reported in

Welcome to May 20, 2026

Actor

Anthropic

Evidence for

Supersedes

A lab will brief central banks on flaws its model found in the financial system

Referenced by

Cite or correct this record

Citation formatCitation text

Imperceptible audio hijacks thirteen speech models
Innermost Loop KB — secondary record
Reporting date: 2026-05-20
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2026-05-20-audio-attacks-hijack-thirteen-models
Claim: Researchers demonstrated imperceptible audio attacks hijacking thirteen audio language models at 79 to 96% success, while Anthropic reversed course and told its security partners it fully supports publishing their findings.
Attributed sources:
- Welcome to May 20, 2026 — https://theinnermostloop.substack.com/p/welcome-to-may-20-2026
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.