EN PT-BR
← Voltar ao feed
medium financial services retail telecommunications energy software

Actor impersonates ~32 AI bot identities (ChatGPT, Claude, Perplexity) to scan sites for exposed secrets, per Cequence

1 fonte(s) · 1 atualização(ões) · publicado 02/11/2026 05:13 · coletado há 1 hora

Cequence reported a scanning campaign in which a single actor rotated through about 32 AI bot identities, including ChatGPT, Claude, and Perplexity crawlers, to probe enterprise websites for exposed passwords, API keys, and cloud credentials. On one October day the activity reached 48% of Cequence-protected organizations that received any AI bot traffic. Cequence observed it from at least late September through late October, targeting financial services, retail, telecommunications, energy, and software organizations. Nearly all requests were blocked or failed, and the findings do not establish widespread credential theft.

Detalhes técnicos

  • The scanner ran from a major cloud provider and impersonated crawlers and assistants tied to OpenAI, Anthropic, Meta, Perplexity, and others.
  • It requested roughly 90 paths tied to exposed secrets: .env files, cloud credential files, private SSH keys, and server environment information.
  • Each source address sent bursts of about 50 to 190 requests in eight seconds before moving on.
  • It first requested random nonexistent paths to identify sites that return responses regardless of whether a resource exists.
  • Tells: it rotated evenly through all 32 identities, claimed competing AI providers within seconds from the same machine, and reused Anthropic's crawler contact address across different Claude identities.
  • The actor used over 300 cloud IPs in one day and nearly 400 in one week; none seen in late September reappeared a month later. Threat feeds flagged many addresses only after the scanner had moved on, so stable request patterns were a better continuity signal than infrastructure.
  • The impersonator generated more than 80% of purported AI bot traffic at several organizations, inflating estimates of legitimate agent adoption.
  • The article says this is distinct from Axios reporting on unvetted agents accessing U.S. bank accounts, which generally involved customer-provided credentials.

Remediação

  • Never grant automatic trust based on a claimed AI affiliation; verify via signed requests, published network ranges, or direct integrations.
  • Continuously assess requested resources and activity patterns, combining identity verification with behavioral enforcement.
  • Remove secrets from web-accessible locations and review exposed staging and administrative systems.

Linha do tempo

  1. gbhackers.com ↗ · pub. 02/11/2026 05:13 · coletado 02/11 06:00 primeiro registro

    Actor impersonates ~32 AI bot identities (ChatGPT, Claude, Perplexity) to scan sites for exposed secrets, per Cequence

    ler a análise desta fonte

    Cequence reported a scanning campaign in which a single actor rotated through about 32 AI bot identities, including ChatGPT, Claude, and Perplexity crawlers, to probe enterprise websites for exposed passwords, API keys, and cloud credentials. On one October day the activity reached 48% of Cequence-protected organizations that received any AI bot traffic. Cequence observed it from at least late September through late October, targeting financial services, retail, telecommunications, energy, and software organizations. Nearly all requests were blocked or failed, and the findings do not establish widespread credential theft.

    Technical Details

    • The scanner ran from a major cloud provider and impersonated crawlers and assistants tied to OpenAI, Anthropic, Meta, Perplexity, and others.
    • It requested roughly 90 paths tied to exposed secrets: .env files, cloud credential files, private SSH keys, and server environment information.
    • Each source address sent bursts of about 50 to 190 requests in eight seconds before moving on.
    • It first requested random nonexistent paths to identify sites that return responses regardless of whether a resource exists.
    • Tells: it rotated evenly through all 32 identities, claimed competing AI providers within seconds from the same machine, and reused Anthropic's crawler contact address across different Claude identities.
    • The actor used over 300 cloud IPs in one day and nearly 400 in one week; none seen in late September reappeared a month later. Threat feeds flagged many addresses only after the scanner had moved on, so stable request patterns were a better continuity signal than infrastructure.
    • The impersonator generated more than 80% of purported AI bot traffic at several organizations, inflating estimates of legitimate agent adoption.
    • The article says this is distinct from Axios reporting on unvetted agents accessing U.S. bank accounts, which generally involved customer-provided credentials.

    Remediation

    • Never grant automatic trust based on a claimed AI affiliation; verify via signed requests, published network ranges, or direct integrations.
    • Continuously assess requested resources and activity patterns, combining identity verification with behavioral enforcement.
    • Remove secrets from web-accessible locations and review exposed staging and administrative systems.