Munich Re Buys At-Bay for $575 Million to Own AI Cyber Underwriting

Munich Re agreed on August 19, 2026 to pay $575 million for At-Bay, the AI-native cyber insurtech. The price is less than half the $1.35 billion valuation At-Bay carried after its 2021 Series D, and roughly 2.1x its $278 million of 2025 gross written premium. At-Bay will be folded into Hartford Steam Boiler rather than kept as an arm's-length capacity relationship, which makes this a clean build-versus-buy verdict.

Key Takeaways

• A repricing, not a growth story. Roughly 2.1x premium against the $1.35 billion valuation from a 2021 Series D that cited 800% year-over-year premium growth at the time.
• At-Bay credits Active Risk Monitoring with ransomware frequency roughly seven times below the industry average, but part of that gap is risk selection at bind rather than a treatment effect of the monitoring.
• Active mitigation has no mechanism against the aggregation tail, the exposure the capacity market priced with $250 million of new non-proportional cyber reinsurance capacity in the first half of 2025.
• US cyber premium fell 7.11% to $9.14 billion in 2024, the first annual decline the NAIC has recorded for the line, while active policy counts held roughly flat at 4.37 million.

The Deal and the Valuation Reset

At-Bay's connection to Munich Re did not start with the announcement. Hartford Steam Boiler has been an At-Bay partner since the company's founding in 2017, and Munich Re Ventures' HSB fund was among the investors backing the 2021 Series D. This is not a reinsurer buying a stranger's book sight unseen. It is a reinsurer that has spent close to a decade underwriting behind At-Bay's own risk selection, converting a capacity-and-minority-stake relationship into full ownership of the underwriting engine, the claims history and the scanning infrastructure.

The terms are straightforward: $575 million in enterprise value, closing expected in the first quarter of 2027, At-Bay sitting inside HSB rather than continuing as an independently branded MGA, on $278 million of gross written premium and $23 million of cyber fee service revenue as of December 31, 2025, with roughly 280 employees.

Neither company's materials dwell on what the price implies. At-Bay closed a $185 million Series D in July 2021 at a $1.35 billion post-money valuation, then added a $20 million extension bringing the round to $205 million and cumulative funding to roughly $272 million. A $575 million sale five years later sits well below the peak the venture market assigned during the 2020-2021 cyber hard market, even as the premium base and market position both grew.

That compression is a data point for every venture-backed cyber MGA still looking for an exit. A platform can grow its book, hold a top-10 position, and still sell for less than 43% of the equity value it commanded when growth capital rather than underwriting profit was the metric investors priced against. Munich Re did not buy the growth story; it bought a book and an infrastructure it has watched perform from the reinsurance side, at a price built on realized economics.

Selection Versus Treatment

At-Bay's model bundles cyber coverage with continuous external vulnerability scanning and pushed remediation guidance, monitoring security posture through the policy term rather than treating it as a one-time questionnaire. Its own published data makes three claims: ransomware frequency roughly seven times below the industry average, a policyholder 25% less likely to experience an incident after resolving a flagged alert, and 80% of policyholders patching a flagged vulnerability within 2.5 months, twice as fast as the general population.

Those figures are the entire commercial argument for the price. They are also where a pricing actuary has to slow down.

The sevenfold comparison is a portfolio-level statistic measured against an outside population that never went through At-Bay's underwriting funnel. Businesses that buy a policy bundled with continuous scanning, and actually engage with the resulting alerts, are plausibly more security-mature to begin with. Part of that gap is therefore a selection effect baked in at the point of sale, not a treatment effect produced by the monitoring.

The 25% figure is closer to a clean causal comparison, because it holds the underlying population constant and compares insureds who resolved a given alert against those who did not act on the same alert. Crediting the full portfolio-level gap to the mitigation product, rather than isolating what alert resolution buys once selection is held constant, is how a rating plan double-counts a benefit that underwriting selection at bind already captured.

That is the actuarial question Munich Re is buying the answer to, and it separates this deal from the alternatives other buyers tested in 2026. Duck Creek bought orchestration vendor Send ten weeks after launching its own agentic platform; Cowbell built OMNI internally rather than acquiring a rival book; nsur.ai prices a copilot at roughly $2 a deal so a carrier never has to choose. Munich Re did none of those. It concluded that six years of watching a proven selection engine from the capacity side was not enough to replicate the data advantage internally.

The Tail Active Mitigation Cannot Reach

Cyber is conventionally reserved as a short-tail line; most claims develop and close within a year or two. That coexists with a tail an individual policyholder's patching speed cannot touch: silent aggregation and systemic-event risk. A single compromised cloud provider, software vendor or managed service platform can produce simultaneous claims across large numbers of otherwise well-secured insureds.

The IAIS's December 2025 report found the sector's aggregate systemic risk profile declined slightly and remains well below banking's, but supervisors continue to track affirmative and silent cyber specifically because a major systemic event, not idiosyncratic ransomware, is the scenario that could reverse the current pricing trend.

Active Risk Monitoring plausibly bends idiosyncratic frequency, the loss driver that appears when one insured misconfigures a server or ignores a phishing warning. It has no mechanism, by design, against a correlated event hitting the entire book simultaneously through a shared vendor dependency. Remediation speed at any single insured does nothing to stop a supply-chain compromise upstream of all of them.

The capacity market prices that gap separately from primary underwriting quality. Lockton Re tracked $250 million of new non-proportional cyber reinsurance capacity entering in the first half of 2025 alone, raised specifically to absorb the aggregation scenarios a bind-time risk score cannot underwrite away.

Ownership does not retire that exposure. It relocates it. As an independent MGA, At-Bay placed capacity with multiple carriers, so whatever margin sat above its retention flowed in part to outside reinsurers. Inside HSB, that premium and retention move through Munich Re's own group retrocession programme instead, which retains more underwriting margin on a book it spent years reinsuring, and takes the aggregation tail directly onto its own balance sheet rather than sharing it across the open market. It does so into a line where premium recorded its first annual decline in 2024 while policy counts held flat, meaning the softening is rate rather than demand.