Neutrinos launches Kamios, governed agentic AI for insurance and banks

Neutrinos said Wednesday it launched Kamios, a governed agentic AI orchestration platform that lets insurers, banks, and financial services firms run AI agents on real cases with every action declared, evidenced, and metered.

Insurers and banks keep stalling agent pilots because a single claim or onboarding case crosses legacy cores, unwritten rules, and human sign-offs. Kamios’s bet is that a signed mandate plus per-case metering and an audit trail is what lets regulated firms ship agents without giving up the control a board has to keep.

On Wednesday, 30 September 2026, Neutrinos announced Kamios. Neutrinos calls itself an AI automation and orchestration company. Kamios, at kamios.ai, is a new Neutrinos product. The company calls it a governed agentic AI orchestration platform for insurance and regulated industries. Agentic, here, means the software can take the next step on a case, not only draft a reply. Orchestration means lining those agents up with the people, the systems, and the decisions a case already crosses. The release says Kamios gives insurers, banks, and financial services firms one place to run agents on real work, across the core platforms those firms already operate. A core platform is the system of record a company already runs, such as the policy system or the claims system. Every action is governed, metered, and attributable. Governed means a rule says what the agent may do. Metered means the cost is counted. Attributable means a reviewer can see who, or what, took the step. The PR Newswire page stamps the release Sep 30, 2026, 10:00 ET, which is 10:00 a.m. Eastern. The dateline is New York and Singapore. The source line is Neutrinos. Canada Newswire carries the same release. Those lines are Neutrinos'.

Neutrinos says companies have shown that agents can perform individual tasks, and that production is where the work stalls. A single case crosses many people, systems, rules, and approvals. Running agents through that case without giving up control, a trail a reviewer can follow, or a cost the company can predict has become a barrier to putting AI into daily operations. The release cites a Gartner prediction that more than 40 percent of agentic AI projects will be canceled by the end of 2027, because of rising costs, unclear business value, and risk controls that are not strong enough. More than 40 percent is more than two projects in five. The citation Neutrinos prints is Gartner, “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027,” 25 June 2025. That prediction is Gartner’s, as the release cites it. It is not a count Neutrinos made.

The client results on the release are Neutrinos’ account of its own programs, and the insurers are not named. At a Tier-1 insurer, straight-through processing on new business moved from 2 percent to 35 percent. Straight-through processing, shortened to STP, means a new-business case finishes without a person stopping it. Two percent is about one case in fifty. Thirty-five percent is about one case in three. At another Tier-1 insurer, a claims program raised automation from 45 percent to 78 percent without replacing the core platform. Forty-five percent is a bit under half. Seventy-eight percent is a bit more than three cases in four. Tier-1, in the release, is Neutrinos’ label for a large insurer. The release does not name either company, and it does not say an outside firm checked the two figures. The subhead says the product was proven in insurance, where straight-through processing on new business rose from 2 percent to 35 percent, and is now available to banks and financial services firms.

Six client programs run on the platform today, Neutrinos says, across insurance and financial services. The work they cover is customer onboarding, claims adjudication, new business issuance, travel claims, and underwriting. Onboarding is taking a new customer on. Adjudication is the decision on a claim. Issuance is putting a new policy in force. Underwriting is deciding whether to take the risk, and at what price. The release does not name the customers on those six programs.

Work enters through any channel: a voice call, a chat, an email, or a system event. A system event is a signal from another computer, such as a claim file arriving. The platform runs the case from start to finish across six governed layers: Declare, Ground, Orchestrate, Act, Govern, and Improve. A team writes the rules it already follows, in plain English, and a manager signs off. That signed text is the mandate: what the agents are allowed to do, and where they have to stop and ask a person. Agents and people share one record of the case. Any model can be used for the task it suits. A model, here, is the AI system that reads and writes. The platform can run entirely inside the customer’s own environment. Authority, the evidence the agent used, and the cost are written down as each decision is made, rather than reconstructed afterward. Those lines are Neutrinos’, on the release.

What the release says changes for each kind of firm. For insurers, the core stays and the throughput changes. Throughput is how much work gets through. New business, underwriting, and claims are supposed to run straight through on the systems already in place. For banks, onboarding, KYC, dispute handling, servicing, and credit decisioning run the same way, with the audit trail written as the decision is made. KYC, know your customer, is the identity check a bank runs before it takes someone on. An audit trail is the record a reviewer can check later. For financial services firms, client servicing, collections, and reconciliation follow the same governed model. Reconciliation means matching two records so they agree. Neutrinos says the cost of the AI becomes a line on the books rather than an estimate, reconciled per case, per mandate, and per team. Those lines are Neutrinos’. The release does not print a price.

Suresh Chandrasekharan, co-founder and chief technology officer, said agents do not fail in the demo. They fail on the third hand-off, when a case crosses a legacy core, a rule nobody wrote down, and a person who has to sign. A hand-off is the moment the case passes from one person or system to the next. A legacy core is an older system the company still runs. He said Neutrinos built Kamios so that the authority an agent acts under, the evidence it acted on, and the cost it incurred are written into every decision as it happens. He said that is what lets a regulated business run agents at scale with the control, the trail, and the economics a board can stand behind. That quotation is his, in the release.

Samik Ghosh, co-founder and chief executive, said Kamios was proven in insurance first, because that is where Neutrinos was already running production work. He said banks and financial services firms are asking for the same measurable results. He said Kamios has been built to help every enterprise that has to run AI where the rules are real. That quotation is his, in the release.

Every Kamios deployment includes Neutrinos engineers, the release says. They build the first workflows alongside the customer’s own team, hand that work over, and move to the next use case. A workflow is the path a case takes. The point, as Neutrinos states it, is that the next workflow starts from what already works, rather than from another discovery phase. Those lines are Neutrinos’.

Kamios is at InsureTech Connect 2026 this week. The conference runs from 29 September through 1 October at the Mandalay Bay Convention Center in Las Vegas. Neutrinos names booth 3018. On 30 September at 12:05 p.m., on the Life, Health and Benefits stage, Chandrasekharan joins senior leaders from two major carriers for a panel titled “What agentic AI actually costs to run: Lessons from insurers in production.” The release does not name those two carriers, and it does not print a time zone on the 12:05 p.m. line.

The about box for Kamios says the product is one governed place to run AI agents on live cases, from claims and underwriting to onboarding and credit decisions, without replacing what is underneath. It says every action is declared under a mandate, evidenced as it happens, and metered per case, so leaders can answer for what their agents do. It says Kamios runs on the customer’s own infrastructure, in the customer’s own region, with the models the customer chooses. It says the product is in production today at Tier-1 carriers, where straight-through processing on new business moved from 2 percent to 35 percent. The body of the release attaches that 2 percent to 35 percent move to one Tier-1 insurer. The about box says carriers, plural. Both wordings are on the release. The about box for Neutrinos says the company connects systems that sit apart and automates insurance work across new business, underwriting, claims, policy servicing, and distribution. It says Neutrinos is trusted by more than 25 insurers across more than 15 countries. Those are company figures. The release does not say all 25 are on Kamios. The media contact is pressrelation@neutrinos.com.

The platform page at kamios.ai/platform, read the same day, spells the six layers out. Declare: equip agents with the knowledge, tools, and permissions a reviewer signs, written in plain language as a mandate. Ground: tie every agent to trusted data, business context, and the documents the work arrives as. Orchestrate: line up agents, people, and workflows on one case. Act: put the work across models, systems, and channels. Govern: make every action observable, every decision something an auditor can check, and every agent accountable. Improve: use tests and live feedback, and send a change through the same gates a software release passes. The page says the platform can run on the customer’s own machines, in a private cloud, in a mix of the two, or as a managed service, with the customer’s models, and with processing that stays in the region the customer’s rules require. It says personal data is masked before a model sees it. It says Kamios is SOC 2 Type II and ISO 27001 certified, and that those audits are repeated every year. SOC 2 Type II is an outside audit of how a company protects data over a stretch of time, not a one-day checklist. ISO 27001 is a standard for how a company runs information security. The page says ISO/IEC 42001, a standard for managing AI systems, is still in progress. Those certification lines are on the platform page. They are not in the 30 September press release. The page does not print a price.

The Kamios homepage, the same day, calls the product one governed execution layer for agents, people, systems, and decisions, and says evidence sits on every action. It says Kamios is new and the foundation under it is not. The page says the product is built on Neutrinos’ existing automation platform, and on years spent inside the workflows insurers and banks already depend on. It lists the work it is aimed at. For insurance: new business and underwriting, claims, service and retention, and risk and compliance. For banking: lending and credit, risk and compliance, customer service, and payments and operations. For financial services: onboarding and verification, risk and compliance, disputes and chargebacks, and payments and operations. A chargeback is a payment a cardholder disputes. A line on the page says a firm can grow the work without adding headcount, that cycle times are measured in minutes, and that accuracy holds under audit. The page does not attach a number, a customer name, or a study to those three lines. The 2 percent to 35 percent figure and the 45 percent to 78 percent figure are on the press release.

In plain terms, Neutrinos said on Wednesday that an insurer, a bank, or a financial services firm can run AI agents on live cases inside the systems it already has. A manager signs the rules in plain English. The authority, the evidence, and the cost are written down as each decision is made. Neutrinos says one Tier-1 insurer’s straight-through rate on new business moved from 2 percent to 35 percent, and that another Tier-1 insurer’s claims program moved from 45 percent automation to 78 percent, without a new core. Those figures are the company’s account of its clients. Six programs are in use, covering onboarding, claims, new business, travel claims, and underwriting. The customer can keep the platform in its own environment and can choose the model. Kamios is at InsureTech Connect in Las Vegas this week, at booth 3018.

The picture is the Kamios platform graphic from the product page. A white stack of slabs rises in five layers, labeled from the bottom: Ground, Orchestrate, Act, Govern, and Improve. Ground carries a database icon. Orchestrate carries connected nodes. Act carries a processor. Govern carries a shield. Improve carries a circular arrow. The title on the card reads “One platform. Six layers of governed work.” The line under it says one agent on its own is a demo, and that the real job takes several agents, the customer’s people, and a dozen systems, under one set of controls, leaving one account of what happened. The footer reads Kamios, then “Declare the mandate. Run the work. Prove what happened.” The graphic draws five of the six layers named in the release. Declare, the first layer in that list, is not its own slab on this card. It is a product diagram. It is not a photograph of an office, and it does not print a date.