{
  "id": 9339197,
  "title": "ZeroDrift launches three models for real-time AI compliance checks",
  "url": "https://urgent.news/2026/09/23/zerodrift-launches-three-models-for-real-time-ai-compliance-checks",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-23T13:00:48.000Z",
  "source": {
    "name": "SiliconANGLE",
    "slug": "siliconangle",
    "url": "https://siliconangle.com/2026/09/23/zerodrift-launches-three-models-for-real-time-ai-compliance-checks/"
  },
  "original_language": "en",
  "account": "ZeroDrift Inc., a startup automating compliance for AI communications, unveiled Anchor 3.0, a trio of compact language models capable of scrutinizing AI-generated messages in real time. These models enforce financial regulations and internal policies, ensuring messages adhere to rules before being dispatched. The family is available via ZeroDrift's Enforcement application programming interface, addressing the challenge of autonomous agents generating large volumes of customer communications faster than human compliance teams can review them.\n\nThe flagship model, Anchor 3.0, detected over 90% of violations in a benchmark using Financial Industry Regulatory Authority rules. It matched the accuracy of OpenAI Group PBC's GPT-6 Astra and Anthropic PBC's Claude Fable 5.1 while running 100 times faster and costing less than one-fifth as much. Benchmarks were based on attorney-labeled data from Surge AI Inc.\n\nFounder and CEO Kumesh Aroomoogan noted that while frontier models enabled building capable agents, the real challenge lies in running them in regulated businesses where every message must comply with rules before transmission. Anchor operates within ZeroDrift's compliance platform, which intercepts and checks communications against regulatory requirements and company policies. The system can flag, rewrite, block, or route problematic messages for human review while documenting decisions for audits.\n\nAnchor 3.0 comes in three versions. Mini, a 9-billion-parameter mixture-of-experts model with 4 billion active parameters, targets high-volume traffic. It runs prebuilt rule packs and flags violations, outperforming Claude Fable 5.1 and GPT-6 Astra in a FINRA test, with fewer false positives. The flagship Anchor 3.0 can apply over 200 prebuilt rules covering various regulations and identify violated lines, rewriting them before message release. Max, a 27-billion-parameter model, handles long documents and attachments, enforcing company policies without additional fine-tuning, utilizing post-training from Google's Gemma E4B and Alibaba's Qwen3.8-27B models. The production release followed an August preview where ZeroDrift described Anchor as a risk-reduction layer, not a guarantee against all violations.",
  "summary": "ZeroDrift Inc., a startup that automates compliance for artificial intelligence communications, today launched Anchor 3.0, a family of small language models designed to check messages generated by AI agents before they’re sent. The company said the models can enforce financial regulations and internal corporate policies while operating fast enough to examine every outgoing message in […] The post…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}