Alex Taylor
Founder
About ModelStrike
ModelStrike exists because AI systems are becoming actors, not just interfaces.
Traditional chatbots mostly produced text. If one was manipulated, the damage was usually limited to what it said. Agents are different.
Agents can:
- Browse
- Execute
- Retrieve
- Modify
- Communicate
- Purchase
- Deploy
- Operate software
When software can decide which actions to take, security testing must account for both application vulnerabilities and model-driven behavior.
A conventional penetration test asks whether an endpoint can be abused. An agent assessment also asks whether the system can be talked into abusing it — by a user, by a document it reads, or by its own chain of decisions.
ModelStrike focuses on that boundary.
Approach
Expert assessments today. Repeatable tooling over time.
ModelStrike begins with hands-on, expert-led security assessments. As testing techniques prove repeatable across engagements, our direction is to turn them into automated AI-security tooling — so teams can keep testing as their agents change.
Team
Who you'll work with
Before you give an AI agent more access, find out how that access can fail.
ModelStrike can evaluate your agent architecture, attack surface, and production security controls.