Attacker spoofed OpenAI and other AI bots to scan for secrets
Original title: Imposter abused trust in OpenAI, Anthropic, and 30 others to hunt for secrets
One actor reached 48% of Cequence-protected organizations that saw AI bot traffic on one October day; nearly all requests failed.
Cequence says a single threat actor impersonated about 32 AI bots, including OpenAI, Anthropic, Meta, and Perplexity, while probing enterprise web servers for secrets. The scanner ignored pages and requested roughly 90 paths associated with passwords, API keys, cloud credentials, and SSH keys across finance, retail, telecom, energy, and software. Activity ran from at least late September through late October 2026 from a major cloud provider. Cequence did not report a breach and said nearly all requests were blocked or failed.
Unconfirmed: Cequence attributes the scanning to one actor but did not report a confirmed breach; the cloud provider is unnamed.
- Bursts sent 50 to 190 requests in about eight seconds, then moved on.
- The same machine claimed OpenAI, Anthropic, Meta, and Perplexity within seconds.
- It used more than 300 cloud addresses in a single day.
- Targets included .env files, cloud credential files, private SSH keys, and environment data.
Impact: Attempts to retrieve passwords, API keys, and cloud credentials
- 1.Verify AI crawlers via published address ranges, signed requests, or direct integrations.
- 2.Flag bots that request credential files and keep secrets out of web directories.
- 3.Replace long-lived local credentials with centrally managed secrets.
- Organizations
- CequenceOpenAIAnthropicMetaPerplexity
From the original article
2 min read at csoonline.com
Attackers are impersonating AI agents from companies including OpenAI, Anthropic, and Meta in a bid to trick organizations into exposing sensitive credentials, according to researchers who observed a single threat actor cycling through dozens of fake bot identities while probing enterprise web servers for secrets.
csoonline.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.