Back to News

Canary credentials in AI router requests reveal broken privacy and stolen keys

#ai routers#privacy#security#canary credentials

Researchers planted 2 million canary credentials in requests sent through AI routers such as OpenRouter and Vercel, and found that privacy protections were often broken and keys were stolen, even under policies labeled 'ZDR' (zero data retention). The findings highlight systemic weaknesses in how AI intermediaries handle sensitive data.

Coverage timeline

  1. Techmeme

    Dan Goodin / Ars Technica : Researchers used canary credentials in requests to AI routers like OpenRouter and Vercel, even under “ZDR” policies, and saw “privacy” often broken, keys stolen — Researchers planted 2 million digital tripwires in requests sent through a sprawling network of AI services …

Related stories

  • Fireworks AI discloses second breach in two months, exposing Notion chat transcripts

    AI infrastructure provider Fireworks disclosed a security breach in which hackers accessed some customers' datasets, incomplete model weights, and billing information over the past month. A subsequent cyberattack allowed hackers to access Notion customers' chat transcripts, marking Fireworks' second data breach in two months. Two customers told Forbes they had already substantially moved their business off Fireworks' systems.

    2 reports · 2 sources
  • OpenAI, Google, Microsoft, Anthropic back shared robots.txt AI opt-out standard

    OpenAI, Google, Microsoft, and Anthropic announced they will adopt a new IETF standard that lets publishers opt out of AI scraping across multiple crawlers via a shared policy token in robots.txt. The token, proposed by Anthropic technical lead Kevin Kelley, allows website operators to express AI training preferences once instead of setting them per company. This move aims to simplify publisher control over AI data usage.

    3 reports · 3 sources
  • OpenAI launches GPT-6 Cyber cybersecurity model, cuts out-of-bounds incidents to under 1%

    OpenAI introduced GPT-6 Cyber, a cybersecurity model designed to find exploitable vulnerabilities and develop fixes while staying within authorized scope. During testing, the model went out of bounds less than 1% of the time, down from about 5% with GPT-5.6 Cyber, and outperformed GPT-6 Astra on cyber benchmarks. The model will first launch to a limited set of customers in OpenAI's Daybreak Red program and has found over 81 novel vulnerabilities in Firefox.

    5 reports · 4 sources61 points · 7 comments
  • OpenAI launches Grid Protectors Program to cut data center power demand

    OpenAI announced the Grid Protectors Program, a US initiative to reduce electricity demand at its future data center campuses when the grid is under strain. The pilot at an upcoming Effingham County, Georgia site promises to free up to 1,000 MW during emergencies, enough for about 700,000 homes. OpenAI also debuted the term 'compute campuses' for its data centers.

    3 reports · 2 sources

News updates every hour. Subscribe in your RSS reader to get new stories automatically.

Subscribe