Reflection AI, which builds open-weight models, launched a Trusted Access Program on 17 November and is starting it with Laser, an autonomous penetration-testing harness that only vetted organizations can use. Laser is “available as a research preview exclusively to commercial partners accepted into the program”, Reflection said.
Our number: Laser was gated 43 days after Reflection introduced Beam, its 501B open-weight model, on 5 October, by our arithmetic from the dates on Reflection’s blog. Beam is an open-weight model; the hacking tool built for Reflection’s agent is gated.
Business Pill · Red Teaming
An evergreen Business Engineer explainer on red teaming, attacking your own AI system before someone else does. It teaches the general idea only and does not discuss Reflection or Laser.
The key insight: Reflection AI introduced Beam, an open-weight model, on 5 October and gated Laser, its autonomous penetration-testing harness, 43 days later by our arithmetic; Laser is available only to commercial partners accepted into its Trusted Access Program, the company said.
What Laser Does
Laser is a penetration-testing harness for Mirror, Reflection’s terminal agent, the company said in an X post and its blog. “One launch command drives a scoped, ledgered engagement”, Reflection wrote.
An executor acts on the target and records what it observes. A separate supervisor reviews that evidence and guides the next turn; “when work stalls, it can direct the executor to delegate independent investigations to workers sharing the same scope and ledger”, the company said.
Reflection said the findings “are verified by attack graphs and captured evidence”. It said Laser was built for “push-button black box assessments of web vulnerabilities and common network penetration techniques”.

What Was Posted on X
Reflection AI’s own launch post on X, 17 November 2026: it introduces the Trusted Access Program and Laser, the autonomous penetration-testing harness for its Mirror terminal agent. This is the piece’s primary for the launch.
A relay post from @choblin29 on 18 November summarising the launch: Laser tests networks, verifies vulnerabilities with captured evidence and delegates to other agents when stuck, and is restricted to vetted commercial partners. The piece checks each point against Reflection’s own blog and Axios.
Who Gets Access
The program limits cyber capabilities to vetted organizations, Reflection first told Axios. The first participants will include enterprises strengthening their own defenses and cybersecurity companies that want to build products on the technology, Axios reported.
Reflection has not detailed its eligibility requirements or how many participants it plans to accept, according to Axios. Vijay Patnaik, Reflection’s chief business officer, told Axios: “The goal is not to gatekeep cyber or AI safety to a small set of authorized organizations”.
Why an Open-Model Lab Gated This
Reflection’s Beam safety post of 5 November said Beam outperformed every Western open-weight model tested on Irregular’s cyber capability evaluations. It said Beam “solved nine of 14 easy challenges”, about 64% by our arithmetic, but completed no end-to-end CyScenarioBench scenarios.
Irregular “assessed that Beam did not materially alter the cyber threat landscape”, Reflection wrote. Laser adds the scoped ledger, the supervisor and the worker delegation around the agent, and that layer is what Reflection chose to gate.
Anthropic and OpenAI run similar trusted access programs, Axios reported. Our coverage of OpenAI’s GPT-6 Cyber and of how Codex found Java4Shell tracks the same shift toward AI-found vulnerabilities.
The Business Lever
The lever is access, not weights. Reflection has built its business around developing open-weight models, Axios reported; Laser goes first to commercial partners it selects, including cybersecurity companies interested in building products using the technology, Axios reported. Our earlier coverage of CISA’s Java4Shell warning and the patch-driven outages described to Congress covers the defender side.
The Structural Read
As we read the launch, Reflection draws the line at the harness, not the model: Beam’s weights are open, while the scoped, ledgered pen-testing layer around its agent is gated.
As we read Axios’s account, the open question is the access rule itself: Reflection has not detailed eligibility or how many participants it will accept.
Vijay Patnaik, Reflection chief business officer, to Axios, 17 November 2026
“The goal is not to gatekeep cyber or AI safety to a small set of authorized organizations”
Three Implications
FOR SECURITY TEAMS Reflection said Laser is now available for selected organizations conducting authorized security testing, and that applicants can apply through its program page.
FOR SECURITY VENDORS Axios reported that the first participants will include cybersecurity companies interested in building products using the technology.
FOR OPEN-MODEL USERS Reflection said it will work with organizations to understand how these capabilities perform in practice and what their limitations are.
The Business Engineer Lens
This story maps onto the Business Engineer framework AI & The Harness Theory.
The framework says: “It is an operating system that wraps your AI — a harness that compresses your thinking, your team’s feedback, and the model’s raw capability into something replicable.”
As we read it, Laser is that harness for offensive security work, and the harness, not the open model, is the part Reflection gates. The framework predates this article and does not discuss Reflection or Laser.
What Is Not Established
Reflection has not published pricing, eligibility rules or a participant count. Axios reported that defenders have hit hurdles with such programs and that access is sometimes difficult unless someone works for a major company. Laser’s own test results were not published in the launch post. We did not contact Reflection.
The Bottom Line
Reflection introduced Beam, an open-weight model, on 5 October and gated Laser, its autonomous penetration-testing harness, 43 days later. Its chief business officer told Axios the aim is to make such tools available to defenders more broadly over time; Reflection has not given a date.
97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Laser and program details are from Reflection AI’s launch blog and X post of 17 November 2026, read at source. Access terms, participant details and the Patnaik quote are as reported by Axios (17 November 2026). Beam’s cyber results are from Reflection’s 5 November safety post. Day counts and the percentage are our arithmetic. We did not contact Reflection.
Sources: Reflection AI, 17 Nov 2026 · Reflection AI on X, 17 Nov 2026 · Axios, 17 Nov 2026 · Reflection AI, 5 Nov 2026









