Safety Classifiers

Safety classifiers are machine learning systems designed to detect and flag harmful, unsafe, or policy-violating content in AI outputs before they reach end users. They function as automated guardrails by evaluating generated text, images, or other media against predefined safety criteria. For AI governance teams, safety classifiers are critical infrastructure for managing model risk, demonstrating due diligence to regulators, and ensuring compliance with content policies without requiring manual review of every interaction.

1 item

Corporate PolicyGlobal2026-07-24

Opus 5 Launches With 85% Fewer Safety Classifier Triggers and a Separate Data Retention Regime, Complicating Enterprise Privacy Programs

Anthropic released Opus 5, a new flagship model that operates under a substantially lighter safety classifier regime than its Fable 5 and Mythos counterparts, with classifiers expected to engage 85% less frequently. The model is exempt from the 30-day data retention policy that applies to other Anthropic models, creating a distinct compliance profile that enterprises must account for separately. Anthropic also introduced an opt-in Automatic Fallbacks feature that silently routes flagged prompts to a less capable model instead of returning an error.

Anthropic Opus 5 Fable 5 Mythos safety classifiers data retention API compliance model release