Harmonic Security Inc., a San Francisco-based company that provides artificial intelligence capabilities for enterprise data protection, announced today it has raised $17.5 million in early-stage funding.
The Series A round was led by Next47 and brought the company’s total raised to more than $26 million, including $7 million during the company’s seed round led by Ten Eleven Ventures in October 2023.
The company has built an approach to data protection for enterprise businesses using AI tools using pretrained, specialized large language models. Instead of strict, rote rules for detecting sensitive data, Harmonic uses LLMs to help classify, label and identify sensitive data before it’s potentially passed along outside an organization to AI-enabled tools.
“Over the past 20 years, complex data classification and labeling exercises and noisy rules have created headaches for security teams, not least a huge burden of meaningless false positives,” said co-founder and Chief Executive Alastair Paterson. “Now, with the rapid adoption of generative AI, renewed data privacy risks mean that legacy data protection tools are not fit for purpose. Harmonic Security is redefining the entire data protection category which gen AI has brought into sharp focus.”
Harmonic sits between employees and AI tools such as OpenAI’s ChatGPT, Microsoft Corp.’s CoPilot for Microsoft 365 and others where it detects when a prompt will be used. It then catches potentially sensitive data that might go into the tool by understanding how the employee might use the AI tool and for what. It can even warn the users that they might be sending sensitive data to an untrusted AI tool and coach the user to provide safe alternatives to help reduce noisy alerts for the security team.
The company said its pretrained LLMs were built from the expertise and experience of dozens of chief information officers and chief information security officers. To train the LLMs, Harmonic used datasets of highly realistic sensitive materials, allowing it to detect all types of sensitive data accurately in milliseconds. This accuracy and low latency mean that it remains virtually invisible to the users of AI tools.
“Traditional data loss prevention methods have proven ineffective, leading CISOs to shift focus toward user authentication, authorization, and visibility,” said Bain Capital CISO Mark Sutton, who recently became a board member at Harmonic. “While specific types of data can be easily protected, the contextual nature of most sensitive information makes binary ‘yes or no’ controls inadequate, creating more operational friction than value.”
The platform is also capable of detecting more than 6,000 different AI tools, including software-as-a-service and AI-enabled web-based apps, helping reduce the hidden overhead of “shadow AI.” This AI usage appears in a back end for information technology and security teams to help identify generative AI risks from employee use.
The company employs 30 people between the United States and the United Kingdom and said the new funding will go to scaling up its marketing efforts and further its engineering vision.