Confident Security emerges as a game-changer in AI privacy, offering end-to-end encryption for foundational AI models. Their product, CONFSEC, ensures prompts and metadata remain secure, addressing growing concerns over data collection by tech giants. Inspired by Apple's Private Cloud Compute system, CONFSEC anonymizes data and employs advanced encryption techniques. The startup's $4.2 million seed funding signals strong investor confidence in their approach. As AI integration raises privacy concerns across sectors, Confident Security positions itself as the "Signal for AI," promising to enable AI adoption without compromising data security. Their solution could unlock markets previously hesitant to embrace AI due to privacy issues, potentially reshaping how we interact with AI in the future.
In an era where artificial intelligence is becoming increasingly integrated into our daily lives, the question of data privacy has never been more pressing. As organizations across healthcare, finance, and government sectors hesitate to fully embrace AI due to data security concerns, a new player has emerged with a promising solution. San Francisco-based startup Confident Security recently came out of stealth mode with $4.2 million in seed funding, positioning itself as “the Signal for AI” – a reference to the widely respected secure messaging platform known for its privacy-first approach.
Confident Security’s flagship product, CONFSEC, offers end-to-end encryption that wraps around foundational AI models, ensuring that prompts and metadata cannot be stored, seen, or used for AI training – not even by the model provider or third parties. “The second that you give up your data to someone else, you’ve essentially reduced your privacy,” explains Jonathan Mortensen, founder and CEO of Confident Security. “And our product’s goal is to remove that trade-off.”
The timing couldn’t be more critical. Tech giants like OpenAI, Anthropic, and Google are increasingly collecting and retaining user data to improve their models or for safety monitoring purposes, creating legitimate concerns for enterprises dealing with sensitive information. Confident Security aims to serve as an intermediary between AI vendors and their customers, including hyperscalers, governments, and enterprises that need stronger privacy guarantees than what’s currently available.
What makes CONFSEC particularly innovative is its architecture, which draws inspiration from Apple’s Private Cloud Compute (PCC) system. According to Mortensen, Apple’s PCC “is 10x better than anything out there in terms of guaranteeing that Apple cannot see your data” when processing certain AI tasks securely in the cloud. Apple’s PCC extends the security and privacy of Apple devices into the cloud, ensuring that personal user data isn’t accessible to anyone other than the user – not even to Apple itself.
Similarly, Confident Security’s system first anonymizes data by encrypting and routing it through services like Cloudflare or Fastly, ensuring servers never see the original source or content. It then employs advanced encryption that only allows decryption under strict conditions.
“So you can say you’re only allowed to decrypt this if you are not going to log the data, and you’re not going to use it for training, and you’re not going to let anyone see it,” Mortensen explained. Finally, the software running the AI inference is publicly logged and open to review, allowing experts to verify its privacy guarantees.
The $4.2 million seed funding round, led by Decibel with participation from South Park Commons, Ex Ante, and Swyx, signals strong investor confidence in Confident Security’s approach. “Confident Security is ahead of the curve in recognizing that the future of AI depends on trust built into the infrastructure itself,” noted Jess Leão, partner at Decibel, in a statement. “Without solutions like this, many enterprises simply can’t move forward with AI.”
The growing concern over AI data privacy extends beyond enterprise settings to consumer applications as well. Recent developments in AI-powered web browsers highlight the potential risks. Take Perplexity’s new AI browser, Comet, which requests extensive permissions to access users’ Google accounts, including the ability to manage drafts, send emails, download contacts, view and edit calendar events, and even access company employee directories. While Perplexity claims much of this data is stored locally, users are still granting the company rights to access and use their personal information, including potentially to improve its AI models.
Signal president Meredith Whittaker recently compared using AI agents and assistants to “putting your brain in a jar,” explaining how these tools often request far-reaching permissions to perform seemingly simple tasks. An AI assistant might need access to your browser (exposing stored passwords and browsing history), credit card information, calendar, and contacts – all to perform a task as basic as making a restaurant reservation. This represents an enormous amount of trust in technology that, as Whittaker points out, is already prone to errors and hallucinations.
The security risks associated with messaging and AI platforms are further highlighted by recent attacks targeting TeleMessage, a Signal clone app used by high-profile government officials and corporations. In July 2025, cybersecurity firm GreyNoise reported multiple attempts to exploit a vulnerability in TeleMessage (CVE-2025-48927) that could allow hackers to access “plaintext usernames, passwords, and other sensitive data.” The U.S. Cybersecurity and Infrastructure Security Agency (CISA) confirmed that hackers are actively exploiting this flaw.
This comes after TeleMessage experienced a significant data breach in May 2025, when attackers stole the contents of users’ private messages and group chats, including those from Customs and Border Protection and cryptocurrency company Coinbase. The vulnerability (CVE-2025-47729) revealed that the TeleMessage archiving backend held cleartext copies of messages from users, severely compromising their privacy.
These incidents underscore the critical importance of robust security measures in communications platforms, particularly those handling sensitive information. As Zack Whittaker, security editor at TechCrunch, recently noted, “AI tools are more and more asking for gross levels of access to your personal data under the guise of needing it to work. This kind of access is not normal, nor should it be normalized.”
For Confident Security, these vulnerabilities in existing platforms represent both a challenge and an opportunity. The company’s approach to securing AI interactions mirrors what Signal has done for messaging – providing a layer of protection that prevents unauthorized access to sensitive data. As Mortensen succinctly puts it: “You bring the AI, we bring the privacy.”
The one-year-old company reports that CONFSEC has already been tested, externally audited, and is production-ready. They are currently in discussions with banks, browsers, search engines, and other potential clients to incorporate their technology into existing infrastructure stacks. Even AI companies themselves might see value in offering Confident Security’s tools to enterprise clients as a way to unlock markets that have been hesitant to adopt AI due to privacy concerns.
As AI continues to evolve and integrate into more aspects of our personal and professional lives, the need for robust privacy protections will only grow more urgent. Companies like Confident Security are leading the charge in developing solutions that allow organizations to leverage the power of AI without compromising on data security. In a landscape where the default seems to be sacrificing privacy for functionality, their approach offers a refreshing alternative – and potentially a blueprint for how we might interact with AI more securely in the future.
The message is clear: as we embrace the capabilities of artificial intelligence, we must be equally vigilant about protecting our most sensitive information. With the right tools and approaches, we can enjoy the benefits of AI without surrendering our privacy in the process.
