June 10, 2026
Delete after reading... maybe
Anthropic requires 30 day data retention for Fable and Mythos
Your chats may stick around for 30 days, and the comments are absolutely spiraling
TLDR: Anthropic says some business users of its most powerful AI models will no longer get true zero-retention, because prompts and replies may be stored for 30 days for safety checks. Commenters are fighting over whether this is a sensible guardrail or a privacy nightmare, especially for confidential work.
Anthropic just dropped a privacy curveball: if businesses were using its highest-powered new models with "zero data retention" turned on, that promise is changing. Prompts and replies for Mythos-class models — including Fable and future similar systems — can now be kept for 30 days so the company can look for abuse patterns like repeated jailbreak attempts, extortion, or spying campaigns. Anthropic says only a tiny approved team can see flagged conversations, everything is logged, and most data gets deleted after the month is up. But the community? Not exactly soothed.
Over in the comments, the mood swung from skeptical to full-on alarmed. One of the spiciest reactions was that "30 days" sounds more like "at least" 30 days, with users zeroing in on the weaselly phrase "almost all cases" and worrying that sensitive work data — even an entire codebase — could be swept in. Another mini-drama exploded around contracts: if you're using an online AI tool while under an NDA, are you already breaking the rules, or has Anthropic just made that risk a lot more obvious? That question had big "office legal team sweating" energy.
Of course, not everyone was clutching pearls. One blunt reply cut through the panic with a brutal shrug: "Then don’t use it." And a pro-safety faction argued the retention is justified if it helps catch bad actors. So yes, the official story is "responsible safety rollout" — but the real show is the crowd splitting into Team reasonable safeguard versus Team absolutely not with my private data.
Key Points
- •Anthropic will retain prompts and outputs from Mythos-class models for 30 days for trust and safety purposes across all platforms where those models are offered.
- •The policy takes effect on June 9, 2026 and applies to covered models such as Claude Mythos 5 and Claude Fable 5, as well as future similarly capable models designated by Anthropic.
- •The change affects certain organizational users with zero data retention configurations, including those using Claude Console, Claude Enterprise, AWS Bedrock, Google Cloud Agent Platform, or Microsoft Foundry with ZDR.
- •Anthropic says the retention is needed to detect misuse patterns that span multiple requests, including best-of-N jailbreaking, espionage activity, and data extortion campaigns.
- •Anthropic says access to retained data is restricted, logged in tamper-proof systems, and followed by automatic deletion after 30 days except for rare safety investigations or legal requirements.