ZeroDrift launches Anchor 3.0, a family of small language models that perform real-time compliance checks on outgoing messages via its Enforcement API.

Anchor 3.0 is positioned as a production-ready platform for financial institutions, emphasizing regulatory alignment and auditable governance as they scale AI communications.

The system runs compliance checks on each outgoing message in about 1.5 seconds through the API, and under 100 milliseconds when self-hosted.

FINRA benchmark methodology and full results are publicly available, with test data created by Surge AI using human-written, attorney-labeled inputs.

In FINRA benchmarking, Mini detects roughly 5% more violations than Claude Fable 5.1 and about 20% more than GPT-6 Astra, while producing fewer than half as many false positives.

Anchor 3.0 Max is designed for processing long documents and enforcing internal policies without extra fine-tuning, making it suitable for attachments and complex tasks.

ZeroDrift notes the market need is growing as autonomous agents generate large volumes of customer communications faster than human reviewers.

A benchmark shows Anchor 3.0 catching 95.5% of FINRA rule violations, outperforming frontier models on recall, precision, and F1 while offering faster and cheaper performance.

The press release provides background on ZeroDrift, its founding team, notable investors, and includes LaunchSquad contact.

Anchor 3.0 can apply extensive prebuilt rules for FINRA, the SEC, and other regulations, identifying violative lines and rewriting them before release.

ZeroDrift previously introduced Guard for Agents, an API service that integrates checks into agent workflows; the August preview framed Anchor as a risk-reduction layer rather than a guarantee of catching every violation.

Regulatory governance is highlighted as essential for scaling AI, with frontier models noted as slower and more expensive for per-message compliance checks.