Keythorn writes liability coverage for companies deploying AI agents, chatbots, and autonomous workflows into production. The company technically assesses AI systems, certifies them against recognized standards, and places coverage intended to close enterprise deals that would otherwise stall . The pitch is aimed at builders whose AI is starting to move money, make decisions, and interact with customers, along with the insurance carriers that need a way to evaluate a risk class most underwriters cannot yet price in-house.
The launch matters because AI deployment is outrunning the insurance market around it, leaving founders exposed when a model gives a wrong answer, a bad recommendation, or triggers a lawsuit. Keythorn's approach involves stress testing a customer's system, mapping how it fails, and then attaching coverage to the specific weak points that assessment surfaces. The company also positions itself as a channel for carriers, sourcing and underwriting AI risk on their paper while early capacity partners help define the coverage box.
Keythorn is built by Casey Tewey and Michael Wong, both from MIT, and is part of a16z Speedrun . Tewey is cofounder and CEO , with Wong as cofounder and CTO, and the team is based in San Francisco. This launch marks the company's move out of stealth and the opening of a waitlist for AI companies seeking assessments and coverage.
Comments (11)
Sign in to join the discussion.
N
Nadia Verloop28d ago
An agent just wired $40k to a phishing wallet at 3am and you're telling me there's finally a policy for that. About time somebody underwrote the vibes.
K
Kwesi Boateng28d ago
The voiceover on that launch video sounds like it was recorded inside a walk-in closet with a sweater over the mic. Fix the room tone, the pitch itself is solid.
T
Tomas Lindqvist28d ago
hot take: the thumbnail is way too dark and the logo mark disappears at small sizes on the timeline. otherwise the color story actually slaps.
R
Reyhan U.28d ago
ok wait, who is the actual counterparty when the LLM hallucinates a refund policy and the customer sues? that's the part I want to see priced.
M
Min-Ji Lee28d ago
We prototyped something adjacent for internal agent workflows at my last job in 2019 and legal killed it in a week. Curious how you're getting past the actuarial cold start.
J
Joy Mukherjee28d ago
Hey, covering agent risk for TechCrunch this week, would love ten minutes on how underwriting works when the insured entity is non-deterministic. DMs open.
B
Bilal Rahmani28d ago
How big is the team? If more than four people shipped this landing page I'm going to be personally offended.
C
Carlotta Vigo28d ago
Tagline is fine but 'the insurance layer for AI' reads like a Notion doc heading. Try: 'When your agent breaks something, we pay for it.'
O
Omar Adib28d ago
Where is the data on incident claims sitting, and are you planning any EU residency story before the first enterprise call asks? Genuinely curious, not gotcha.
P
Pratik Shinde28d ago
Distribution question nobody's asking: are you selling this to the AI vendor, the enterprise buyer, or bolting onto the agent framework itself? Wildly different GTMs.
H
Halvor Strøm28d ago
Every abstraction layer eventually needs an insurance layer. You are early, which is either genius or two years too soon, and both look identical from here.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Keythorn
The insurance layer for AI
The insurance layer for AI
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Keythorn writes liability coverage for companies deploying AI agents, chatbots, and autonomous workflows into production. The company technically assesses AI systems, certifies them against recognized standards, and places coverage intended to close enterprise deals that would otherwise stall . The pitch is aimed at builders whose AI is starting to move money, make decisions, and interact with customers, along with the insurance carriers that need a way to evaluate a risk class most underwriters cannot yet price in-house. The launch matters because AI deployment is outrunning the insurance market around it, leaving founders exposed when a model gives a wrong answer, a bad recommendation, or triggers a lawsuit. Keythorn's approach involves stress testing a customer's system, mapping how it fails, and then attaching coverage to the specific weak points that assessment surfaces. The company also positions itself as a channel for carriers, sourcing and underwriting AI risk on their paper while early capacity partners help define the coverage box. Keythorn is built by Casey Tewey and Michael Wong, both from MIT, and is part of a16z Speedrun . Tewey is cofounder and CEO , with Wong as cofounder and CTO, and the team is based in San Francisco. This launch marks the company's move out of stealth and the opening of a waitlist for AI companies seeking assessments and coverage.
Comments (11)
An agent just wired $40k to a phishing wallet at 3am and you're telling me there's finally a policy for that. About time somebody underwrote the vibes.
The voiceover on that launch video sounds like it was recorded inside a walk-in closet with a sweater over the mic. Fix the room tone, the pitch itself is solid.
hot take: the thumbnail is way too dark and the logo mark disappears at small sizes on the timeline. otherwise the color story actually slaps.
ok wait, who is the actual counterparty when the LLM hallucinates a refund policy and the customer sues? that's the part I want to see priced.
We prototyped something adjacent for internal agent workflows at my last job in 2019 and legal killed it in a week. Curious how you're getting past the actuarial cold start.
Hey, covering agent risk for TechCrunch this week, would love ten minutes on how underwriting works when the insured entity is non-deterministic. DMs open.
How big is the team? If more than four people shipped this landing page I'm going to be personally offended.
Tagline is fine but 'the insurance layer for AI' reads like a Notion doc heading. Try: 'When your agent breaks something, we pay for it.'
Where is the data on incident claims sitting, and are you planning any EU residency story before the first enterprise call asks? Genuinely curious, not gotcha.
Distribution question nobody's asking: are you selling this to the AI vendor, the enterprise buyer, or bolting onto the agent framework itself? Wildly different GTMs.
Every abstraction layer eventually needs an insurance layer. You are early, which is either genius or two years too soon, and both look identical from here.