Angle Health raised $600M to underwrite small-business plans with AI
Two funding rounds and a policy fight today: an AI-native insurer takes on the small-group market, Europe's own AI firms tell Washington to stop pretending the slowdown is about safety, and the data-loss-prevention business gets rebuilt for agents.
Angle Health raised $600 million led by Vitruvian Partners at a $2.7 billion valuation, and the pitch is narrower than "AI in healthcare": it is algorithmic underwriting for the employers nobody wants to cover. The company sells full-stack health plans to small groups — HIT Consultant reports it targets employers with as few as two lives — and replaces the manual actuarial review that agents and brokers have run on spreadsheets for decades with a model that prices risk from payroll and HR data the company already integrates with. TechCrunch reports it serves more than 5,000 businesses and is profitable. Those last two numbers are the reason this round exists. The first generation of venture-backed insurtechs mostly died chasing unprofitable individual ACA exchange lives or bidding too aggressively on Medicare Advantage pools, and a profitable book in small group is rare enough that Vitruvian paid a multiple for it.
What is actually being automated here is judgment, not paperwork. Small-group underwriting is unattractive precisely because the per-case economics are bad: too few lives to pool risk, too much variation to price by hand, so carriers either decline the group or quote something the owner rejects. If a model can price a five-person company in minutes and still hold a loss ratio, the addressable market is every employer that currently gets a broker's worst rate. That is also the risk: underwriting is an adversarial loop, employers and brokers learn the model's edges, and an insurer that cannot explain a rate cannot defend it in a renewal. ProPublica-style audits of algorithmic pricing start wherever denial rates diverge by ZIP code, and nobody has published that distribution yet.
Europe's leading AI companies accused their US rivals of using safety as a competitive weapon, rejecting the labs' calls to slow the pace of frontier model development. Mistral said in a statement that "some incumbents are using this moment to consolidate their market position, pushing for regulation designed to favour them over competitors," Reuters reports. Proton's chief operating officer Raphael Auphan called the push "totally self-serving," adding that the US labs "just want to preserve a dimension of dependency on their service." Germany's AI professor Kristian Kersting called Dario Amodei's proposal — voluntarily coordinate on safety standards and submit to embedded evaluators in exchange for an antitrust waiver — "regulatory capture," and French finance minister Roland Lescure said he can "clearly see their self-interest" in making everyone behind them slow down. The rebuttal lands the same week Anthropic named Accenture as its first embedded evaluator; we laid out what Amodei was actually offering in Amodei's pacing plan puts outside auditors inside Anthropic. The uncomfortable part for Europe is that the accusation is structural rather than moral: a continent that runs on US models has no leverage over their training schedules either way, which is why Lagarde spent Monday arguing Europe has to build its own compute before it can have an opinion.
MIND raised $72 million to rebuild data loss prevention for a world where the thing moving your data is an agent, not an employee. Crosspoint Capital Partners led the Series B, with YL Ventures and Paladin Capital joining; the round follows a $30 million Series A roughly a year ago and brings total funding to $112 million, and Calcalist reports it values the company near $300 million. MIND says revenue grew more than 17-fold over the past year and its customer count eightfold, on a platform that watches SaaS apps, generative AI tools, agents, endpoints, file shares and email, classifies the content, and blocks or remediates in real time. It also started selling autonomous agents that do the routine triage work of running a DLP program — the vendor's own product line answering the same question enterprise security teams have been asking since researchers used Claude to walk into OpenAI's internal repositories: who reviews what the agents did.
What to watch: whether Angle Health publishes loss-ratio or denial-rate data as it scales, and whether Brussels adopts the "self-serving" framing when von der Leyen's frontier-lab talks actually convene.
If a model prices your company's health plan in minutes, what would you want to see before you trusted the number — an explanation, an audit, or the right to appeal? Tell us in the comments.
