# ZeroDrift applies small language model to prevent AI-generated compliance violations > Source: > Published: 2026-08-11 13:00:35+00:00 ### Exclusive: ZeroDrift applies small language model to prevent AI-generated compliance violations “This investment is guaranteed to return 12% annually.” A claim like that in an email from an investment adviser is a regulatory disaster. Regulations prohibit financial firms from promising returns and require a proviso that past performance doesn’t guarantee future results. In the age of artificial intelligence, fewer communications are going through human hands, and the volume generated by AI agents makes manual reviews impractical. That’s a problem AI compliance startup [ZeroDrift Inc.](https://zerodrift.com/) wants to solve. The company today launched Command, a control plane that automates detecting and correcting regulatory violations in human and AI-generated communications before messages are delivered. Command applies corporate policies and industry regulations to emails and AI agent outputs, flagging such problems as promissory claims, unauthorized disclosures and references to material nonpublic information. ZeroDrift positions the platform as an alternative to conventional compliance systems that review communications after they have already been sent or require time-consuming human review. Command intercepts communications in real time and checks them against a database of regulatory requirements and company policies, said Chief Executive Kumesh Aroomoogan. The system can flag potential violations for review or optionally rewrite the content to be compliant. The distinction is becoming more important as companies deploy AI agents capable of autonomously communicating at high volumes. A single agent can generate thousands of messages in seconds. Financial regulators generally hold companies responsible for AI-generated communications just as they would for messages written by employees, Aroomoogan said. Command consists of four components. Policy Studio converts regulatory documents and internal policies into enforceable rules. Anchor is a small language model that analyzes content for potential violations. Guard applies the policies at the point where communications enter or leave an organization, and Command records the resulting decisions and supporting explanations. The platform can pass, warn, repair or block a message. ZeroDrift automatically rewrites content generated by AI, but it doesn’t change messages written by employees. A human-authored message can instead be blocked, approved by a compliance officer or returned to the author with suggested changes. Command also includes a sandbox in which developers can test messages and see identified violations, proposed corrections and differences between the original and revised text. Its Activity Center provides an audit trail showing the content flagged, the rules invoked and the final disposition. Customers can access the service through an application programming interface and associate individual API keys with specific regulations and internal policies. ZeroDrift also integrates with Microsoft Corp. and Google LLC email systems as an SMTP relay, enabling it to inspect inbound and outbound communications. ### Performance boost Aroomoogan said the company combines deterministic checks, including keywords and pattern matching, with an open-source model post-trained on regulatory data and communications labeled by attorneys. A rewritten message is rechecked before being released to reduce the chance that a correction introduces another violation. ZeroDrift claims the preview version of Anchor 3.0 detects at least twice as many violations as general-purpose frontier models. However, the company has not yet released detailed detection or false-positive results. “We don’t want to release the numbers yet because it’s still in a research phase and those numbers could change,” Aroomoogan said. The company said it is currently evaluating the model against a dataset of 100,000 human-annotated emails. The specialized model is also intended to reduce the expense and delay associated with applying general-purpose large language models to inspection tasks. Aroomoogan said some early approaches relying on Anthropic PBC’s Claude took more than 10 seconds to detect and rewrite each violation. By applying a specialized small language model, ZeroDrift was able to reduce that latency to milliseconds, Aroomoogan said. “We’re definitely under one-second latency in terms of detecting and rewriting messages,” he said. ZeroDrift cautions that Command is a risk-reduction layer rather than a guarantee that every violation will be caught. “We’re not guaranteeing anything,” Aroomoogan said. “At the end of the day, we’re saying that you can use ZeroDrift as a potential risk tool.” The company plans to extend coverage to WhatsApp, SMS, Slack and Microsoft Teams. Pricing is based on the number of input and output tokens processed, but the company declined to provide details. ##### Image: [Pixabay](https://pixabay.com/illustrations/data-privacy-security-binary-code-9611620/) # A message from John Furrier, co-founder of SiliconANGLE: Support our mission to keep content open and free by engaging with theCUBE community. **Join theCUBE’s Alumni Trust Network**, where technology leaders connect, share intelligence and create opportunities. **15M+ viewers of theCUBE videos**, powering conversations across AI, cloud, cybersecurity and more** 11.4k+ theCUBE alumni**— Connect with more than 11,400 tech and business leaders shaping the future through a unique trusted-based network. # Are you AWS customer? Support SiliconANGLE Financially by buying your AWS services from our Marketplace portal page and links. **About SiliconANGLE Media** [SiliconANGLE](https://cts.businesswire.com/ct/CT?id=smartlink&url=https%3A%2F%2Fsiliconangle.com%2F&esheet=54119777&newsitemid=20240910506833&lan=en-US&anchor=SiliconANGLE&index=9&md5=646b1b564e2259100a2b8638aab0a552), [theCUBE Network](https://cts.businesswire.com/ct/CT?id=smartlink&url=https%3A%2F%2Fwww.thecube.net%2F&esheet=54119777&newsitemid=20240910506833&lan=en-US&anchor=theCUBE+Network&index=10&md5=7de2a85f95ab4a4a495cede20b8cb1da), [theCUBE Research](https://cts.businesswire.com/ct/CT?id=smartlink&url=https%3A%2F%2Fthecuberesearch.com%2F&esheet=54119777&newsitemid=20240910506833&lan=en-US&anchor=theCUBE+Research&index=11&md5=7bb33676722925eb57d588ec343e4f6f), [CUBE365](https://cts.businesswire.com/ct/CT?id=smartlink&url=https%3A%2F%2Fwww.cube365.net%2F&esheet=54119777&newsitemid=20240910506833&lan=en-US&anchor=CUBE365&index=12&md5=d310fb35919714e66ad8d42c9c0c1bc6), [theCUBE AI](https://cts.businesswire.com/ct/CT?id=smartlink&url=https%3A%2F%2Fwww.thecubeai.com%2F&esheet=54119777&newsitemid=20240910506833&lan=en-US&anchor=theCUBE+AI&index=13&md5=b8b98472f8071b23ebb10ab9a8dd0683)and theCUBE SuperStudios — with flagship locations in Silicon Valley and the New York Stock Exchange — SiliconANGLE Media operates at the intersection of media, technology and AI. Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.