Toronto-based AutoAlign has spun off from its parent company Armilla AI and has launched a firewall solution aimed at making large language models (LLMs) safer.
According to a spokesperson for AutoAlign, the new spinout is being led by Dan Adamson as interim CEO and co-founder, and Rahm Hafiz as CTO and co-founder. Adamson will continue to work at Armilla AI, but Karthik Ramakrishnan, Armilla AI co-founder and chief product officer, will take over the role of CEO. Hafiz will serve at AutoAlign on a full-time basis, moving from his role as CTO of Armilla AI.
Founded in 2020, Armilla AI develops risk assurance products for artificial intelligence (AI) models. The startup has some notable backers, including deep learning pioneer Yoshua Bengio, Y Combinator, and Two Small Fish Ventures, among others. Armilla most recently closed $6 million CAD in seed financing in February, which brought its total funding to $7 million CAD.
Adamson told BetaKit that over time, Armilla AI grew to have two businesses operating under one roof: insurance and security. By separating its security arm, Armilla AI will now focus on the insurance component of its business.
“As the AI market exponentially expanded, the company was advised to avoid any potential conflicts of interest—having one company insure models that they also secured—which is why AutoAlign spun off,” Adamson explained. “Now two companies can thrive and even work together.”
Alongside the spinoff, AutoAlign is announcing the launch of its flagship LLM firewall security solution, called Sidecar. Sidecar works by interacting with LLMs to protect against faulty AI incidents, such as hallucinations, biases, gender assumptions, data leaks, and jailbreaks. The company claims Sidecar can reject problematic responses and instruct models to make corrections.
Armilla AI first revealed it was working on AutoAlign in June 2023. The project was described as a low-code, web-based platform aimed at reducing hallucinations and harmful responses from LLMs.
Adamson explained that within the first iteration of AutoAlign, Armilla AI built out sub-solutions such as AutoGuard, aimed at protecting organizations against AI malice, and AutoTune, which works to eliminate biases. “These solutions analyzed generative AI responses but did not interact with the model directly,” Adamson said.
Adamson told BetaKit AutoAlign decided to “revamp” its technology by creating Sidecar.
RELATED: Y Combinator and Yoshua Bengio-backed Armilla AI secures $6 million CAD in new funding
“By the time companies build AI with the proper security guardrails, they’re immediately obsolete,” Adamson said in a statement. “That’s why we built our sophisticated AI firewall solution—Sidecar—so enterprises have LLM security that is comprehensive, nimble, and always evolving.”
AutoAlign has also signed a cooperation agreement with KPMG in Canada. The company says Sidecar is already being used in KPMG’s internal systems. KPMG is also now including AutoAlign as an option to its compliance layer for its Trusted AI framework.
“KPMG’s Global Trusted AI framework and approach to designing, deploying, as well as managing AI models and tools allow us to deliver powerful, trustworthy solutions to clients,” said Ven Adamov, partner and national leader of KPMG in Canada’s Risk Analytics practice and Trusted AI co-leader.
“AutoAlign shares our goal of integrating generative AI into businesses in a responsible manner without compromising the effectiveness of this transformative technology,” Adamov added.
UPDATE (04/23/2024): A previous version of this story stated that Dan Adamson would remain CEO of Armilla AI. Additional information from the company confirmed that Karthik Ramakrishnan will replace Adamson as CEO.
Feature image courtesy of AutoAlign.