As banks experiment with and pilot generative AI, they have a growing need to put guardrails around those systems, to address the models’ propensity to hallucinate, to pick up the wrong information and to perpetuate biases, as well as to comply with security and privacy rules and norms. The latest companies to address this demand are Google and Corridor Platforms.
The founders of Corridor Platforms, which is based in Haworth, New Jersey, spent most of their careers at American Express, working in risk, analytics and governance.
“We started the company seven years ago to focus on how banks could use traditional AI, in a much more responsible manner, and start taking the advantage of some of the new technologies and converting them into proficiencies and customer journeys,” Manish Gupta, Corridor Platforms’ CEO, told American Banker. The company has been working with two Tier 1 banks, one for three years and one for five years, and some smaller banks. (Corridor did not name any of the banks.)
“One of the large hurdles is, how do you govern and control the new risks that come up with generative AI, as well as manage the traditional biases that are already there in regular AI?” Gupta said. Corridor co-developed a model management platform with a large bank, then developed a generative AI version, he said. This is called GenGuardX.
The same bank was working with Google Cloud. “Both of us were helping the bank think through, how do you get something into production?” Gupta said. Corridor also works with other cloud computing providers like Amazon Web Services and Microsoft Azure. “We are not hostage to any environment,” Gupta said, adding banks often use the technology on premises.
Google Cloud provides infrastructure, foundation models and “a ton of off the shelf guardrails and controls that financial institutions can adopt and implement,” Toby Brown, managing director of global retail banking solutions at Google Cloud, told American Banker.
For any given model, a bank will typically have 200 to 300 pages of documentation, Brown said. “There’s also this constant monitoring and testing of everything that you’re doing with that model and ensuring that that’s staying compliant with all your internal processes and risk and regulatory requirements as a bank,” he said. “So while we provide all of this rich tooling and capabilities to help banks do that, Corridor comes in over the top to help accelerate the bank’s work of actually then getting that into production. It gives them this workbench and helps them run through all of that in a way that increases their time to value.”
The danger of a
“From a Google perspective, hallucination is one of the default concerns for customers when they look at generative AI,” Brown said. “And we’ve been very focused on developing controls and guardrails and techniques that you can use to mitigate and minimize that hallucination, whether that’s offering capabilities to ground the AI output in your enterprise data; you can also ground it in Google Search. These techniques help our banking customers ensure that the output they’re getting is accurate and repeatable enough to work for the given use cases.”
Corridor Platforms tracks and validates that the controls are working, Brown said.
With Google, “you have a very established company asserting their quality controls on it, as opposed to open source LLMs or LLMs that you do not know much about,” Gupta said. “We give you a set of tools in the platform to evaluate what are the risks in any LLM, including for toxicity or hallucination.”
Corridor’s software can also help with external guardrails — finding and stopping inappropriate model output before it reaches employees or customers.
The partnership announcement got some early positive reactions from the industry.
“With GenGuardX, it looks like banks can scale generative AI confidently while ensuring compliance, risk mitigation and efficiency,” said Suresh Renganathan, chief technology officer at Teachers Federal Credit Union, in a LinkedIn post. “Looking forward to seeing its impact.”