Iridius is building compliance infrastructure designed to keep AI agents operating within deterministic guardrails in regulated environments [Abstract image from Adobe Stock]
But most of these solutions are vertical, targeting a specific use case with a specific agent and model. Accenture is aiming to flip the script with a recent investment in Iridius, a small Seattle startup focused on building horizontal compliance infrastructure.
A vision of thousands of documents, one knowledge engine
The Iridius website defines its approach as “auto policy execution” involving three core steps: transforming regulations into machine-readable compliance logic, orchestrating compliant workflows and generating continuous evidence so every action is traceable and auditable. The scale of the problem those steps address is considerable. Regulated companies frequently operate under thousands of SOPs, policies and work instructions. On top of that are government regulations that vary by country. “There can literally be thousands of documents that you need to make sure you’re following and staying compliant with, that can be reviewed and audited,” Michel said. Iridius built a knowledge engine that ingests those documents and converts them into structured, machine-readable logic that can be embedded directly in workflows.
When agents improvisation isn’t wanted
While converting SOPs to machine-readable logic addresses the input side of the problem, the output side has its own set of hurdles. GenAI agents are built to reason, adapt and improvise through next-token prediction. The approach can create friction in regulated workflows. “There’s a tension between probabilistic and deterministic solutions. AI agents tend to be more probabilistic,” Michel said. “They’re programmed to reason, adapt, draw their own conclusions and generate outputs. But regulated processes require a much more deterministic workflow.”
Iridius addresses that tension partly through guardrails that recognize when an agent has reached the boundary of what it can do autonomously. “Part of what these systems need to do is recognize when they need to stop and wait for an intervention, a review and approval from the human side, before they’re allowed to proceed,” Michel said. Accenture calls the principle “human in the lead.”
Platform pedigree meets pharma gravity
The ethos to put humans in the lead is implicit in the experience of Iridius’ team. For instance, CEO Mike Kropp spent over two decades at Microsoft and AWS, where he led engineering teams and database migration programs. Co-founder Alistair Lowe-Norris, the company’s Chief Responsible AI Officer, spent 23 years at Microsoft, where he led responsible AI initiatives and authored the company’s Cloud Adoption Framework. The company, which recently raised $8.6 million in seed funding led by Chalfen Ventures (with Accenture Ventures participating), has also assembled an advisory board of former pharma CIOs from Merck, Pfizer, J&J, Novartis, Allergan and Bayer, including Clark Golestani, who spent 24 years at Merck and served as its global CIO and President of Emerging Businesses.
Part of the promise of AI in regulated industries like pharma is cost benefits through efficiency gains. While AI has fueled considerable gains in workflows involving, for instance, document summarization and rote tasks, Michel acknowledged there is considerable work remaining in bending the drug development cost curve.
AI, ironically, could cause more workflow pileups as it accelerates drug discovery, which increases the volume and complexity of clinical trials downstream. “The volume of clinical trials, and the complexity of those trials, is going to continue to increase,” Michel said.
Michel pointed to manufacturing (batch release, deviation management, Corrective and Preventive Action), pharmacovigilance and regulatory submissions as areas where compliance automation could have immediate impact. “It wasn’t just applicable to a specific part of the value chain,” he said. Iridius plans to expand beyond life sciences into financial services and other regulated industries.
How does the consulting giant envision working with Iridius over time? Accenture sees compliance as the layer that connects its broader enterprise AI adoption practice (security, data access, system integration, workforce redesign) to the specific regulatory demands of life sciences clients. “We are largely services,” Michel said. “Together, we’re able to have a discussion with a client about how Iridius’ product could be adopted to drive a solution and a business outcome while Accenture helps rethink the processes, rethink the workforce, integrate the tech into the existing landscape, and make sure data is flowing.” As Michel put it: “Everybody knows how to deploy a chatbot to an employee’s desktop. But enterprise adoption of AI solutions is a very different problem.”
