HomeIntelligenceBrief
BREACH BRIEF🟠 High ThreatIntel

Poison Claude: Fraudulent ‘Cheap AI’ Service Intercepts Every Prompt Sent to Anthropic’s Claude

Okta researchers discovered Poison Claude, a black‑market service that sells cheap access to Anthropic’s Claude by routing queries through its own servers, exposing every prompt. This creates a data‑exfiltration risk that challenges SOC 2 confidentiality controls and privacy compliance.

LiveThreat™ Intelligence · 📅 August 07, 2026· 📰 fortra.com
🟠
Severity
High
TI
Type
ThreatIntel
🎯
Confidence
High
🏢
Affected
2 sector(s)
Actions
3 recommended
📰
Source
fortra.com

Poison Claude: Fraudulent “Cheap AI” Service Intercepts Every Prompt Sent to Anthropic’s Claude

What Happened — Researchers at Okta uncovered “Poison Claude,” a black‑market service that sells discounted access to Anthropic’s Claude model by abusing AWS Bedrock free‑credit programs. Victims pay with cryptocurrency, receive a forged API key, and have their development tools pointed at Poison Claude’s servers, which relay queries to the real Claude and return the answer.

Why It Matters for Compliance & Audit Readiness

  • The service acts as a man‑in‑the‑middle for AI prompts, creating a data‑exfiltration risk that directly challenges SOC 2 CC6 (Confidentiality) and privacy‑related controls.
  • Continuous‑compliance programs must capture third‑party API usage and maintain an auditable trail of who accesses external AI models and what data is transmitted.
  • Verisq’s CookiePLUS privacy capability helps map consent, data‑handling policies, and DSAR readiness to SOC 2 CC6, providing evidence that your organization controls what external AI services can see.

Who Is Affected

  • Technology & SaaS firms that embed generative AI (e.g., software developers, fintech, health‑tech).
  • Any organization that integrates Anthropic Claude or similar LLMs via cloud marketplaces.

Recommended Actions

  • Inventory all AI model integrations and verify that API endpoints point to official provider endpoints.
  • Enforce strict access‑control policies for API keys and require multi‑factor authentication for any credential provisioning.
  • Incorporate AI‑prompt handling into your privacy impact assessments and document consent/DSAR processes.

Source: Fortra Blog – Beware cut‑price AI services that read your every word

Technical Notes

  • Attack vector: exploitation of AWS Bedrock free‑credit program via fraudulent accounts; malicious proxy intercepts API traffic.
  • No CVE is associated; the risk stems from third‑party service abuse and man‑in‑the‑middle interception of data.
  • Data types at risk include proprietary code, business documents, API keys, and personally identifiable information.

Source: same as above

📰 Original Source
https://www.fortra.com/blog/beware-cut-price-ai-services-read-your-every-word

This LiveThreat Intelligence Brief is an independent analysis. Read the original reporting at the link above.

From the Verisq platform · PrivacyOps · CookiePLUS

A privacy incident is a question about your consent record.

CookiePLUS and Verisq AI Trust Operations keep consent, DSAR, and data-handling evidence continuously ready — so a data-exposure event finds you prepared, not scrambling.

See how Verisq AI Trust Operations handles privacy →