Reflection AI has launched a Trusted Access Program that gives vetted organizations early access to selected AI capabilities, starting with Laser, an autonomous penetration-testing tool. Laser is available only as a research preview to commercial partners accepted into the program, the company said on 17 November 2026.
Quick Take
- What: Reflection’s new Trusted Access Program, beginning with cybersecurity.
- First tool: Laser, an autonomous penetration-testing harness for Mirror, Reflection’s terminal agent.
- Who gets it: selected organizations doing authorized security testing; access is a research preview.
- Why gated: Reflection says advanced cyber capabilities are spreading and it wants to give defenders tools that “decrease marginal risk.”
- Bigger picture: part of Project Iris, Reflection’s initiative on safety research, benchmarks and mitigations.
In plain English
A penetration test is a hired attack: security experts try to break into a company’s own systems to find weak spots before criminals do. Laser lets an AI agent run such tests largely on its own. Because the same tool could help real attackers, Reflection is not releasing it publicly. Only screened organizations, testing systems they are allowed to test, can use it.
What happened
Reflection announced the program in a blog post. “Today, we’re introducing Reflection’s Trusted Access Program, a program that gives vetted organizations access to selected AI capabilities and tools for defensive cybersecurity,” the company wrote. “We’re starting with Laser, a tool that lets security teams run autonomous penetration tests.” Axios first reported the gated launch on 17 November.
Background
Frontier labs are increasingly gating their strongest cyber capabilities. Anthropic limits Claude Mythos 5.2’s cyber capabilities to vetted organizations, and OpenAI’s GPT-6.2 system card puts its newest models at its Critical threshold for cybersecurity. At the same time, attackers are using autonomous tools, as the FBI’s recent ARTEX warning showed. Reflection, best known for its open Beam model, is now applying the same gated approach.
How it works
- Scoped engagements: one launch command starts a test limited to an agreed scope and recorded in a shared evidence ledger.
- Executor and supervisor: an executor acts on the target and logs observations; a separate supervisor reviews the evidence and decides the next step.
- Delegation: when work stalls, the supervisor can hand independent investigations to worker agents sharing the same scope and ledger.
- Evidence: findings are verified with attack graphs and captured evidence, so teams can see what the agent tried and why.
- Focus: push-button black-box assessments of web vulnerabilities and common network penetration techniques.
Why it matters
For security teams, autonomous testing could make frequent penetration tests affordable, instead of an annual exercise.
For AI developers, the program is another sign that trusted-access tiers are becoming the norm for dual-use capabilities.
For policymakers and auditors, gated programs raise practical questions about who is vetted, how use is monitored and what happens when access is misused.
What’s still open
- How many organizations are admitted and what the vetting criteria are in practice.
- Independent results on Laser’s accuracy and false positives.
- Which capabilities Reflection plans to add to the program next.
Sources
- Reflection AI, “Launching Reflection’s Trusted Access Program, starting with cyber,” 17 November 2026
- Axios, report on Reflection’s gated cyber tools, 17 November 2026
i10x Perspective
Cyber is where the “release everything” and “lock it down” camps meet. Even a lab known for open models now gates its attack tooling, which suggests trusted access will be standard for any agent that can break into systems. More in Security & Defense.



