The record · five incidents · 2026
In 2026, software started breaking into software. This is the public record, and what to do about it.
Scroll to readHall of Fame researchers, Black Hat speakers and OWASP chapter leads. The same team now reads your product first.
Every entry links to the company's own disclosure. Read it as a briefing, then see how MATT closes each gap.
What changed in 2026
Firefox vulnerabilities Mozilla fixed in two weeks with one model.
Every company on the record had a security team, scanners and a pentest on file. The missing piece was a loop that knew what had become exploitable since.
Software now changes faster than humans can review it, and AI has changed the economics of finding weaknesses in it. Security is still split across tools, teams and point-in-time processes. MATT sits across that loop: it observes what changes, proves what is exploitable, helps your team fix it, and remembers the fix as a control on every release.
Claude Opus 5 turned an OpenAI forum bug into a pull request inside its own code.
On patching: “In the old days you had a week or two. Now it’s got to be like minutes.”
Enterprises buying token-based access to frontier models are paying to expose their intellectual property.
“Nothing that was deployed seven or 10 years ago is prepared to handle AI at machine speed.”
“Most companies have some level of gaps. Even if they are sophisticated and spending a lot.”
Attackers used AI agents to plan, build and run a mass credential-harvesting campaign in under six hours.
“We’ve given AI agents the crown jewels to everything. It’s going to become the No. 1 attack vector.”
“People are going to be able to have ongoing, persistent attacks on you.”
“Organisations should assume that agentic AI systems may behave unexpectedly and plan deployments accordingly.”
A public surface, a capable model, and a check nobody had specified. MATT writes the checks, in plain English, and tests your product against them.
Every public incident this year reduces to a check nobody had written down. We write it as a one-line rule in plain English, such as "no user can read another user's order", and test your product against it.
Findings are ordered by what they cost the business, not by CVSS score. Each one states whether the outcome is an invoice, a disclosure, or downtime, so leadership can decide in one read.
Anyone can create an account, so we start there: every endpoint, token and workflow a new user can touch, mapped before the first test runs.
Generated code passes review and still omits the ownership check or the rate limit. Rules run against the behaviour of the deployed product, so it does not matter who or what wrote the code.
MATT is the AI Security Supervisor for your software lifecycle. It observes what changes, proves what is actually exploitable, helps your team fix it, and remembers the fix as a control. Observe, prove, fix, remember: this is ExploitOps.
The audit works from read access to your repository, your deployment and the tools you already run: scanners, pentest reports, evals, observability. GitHub access is enough to start. The first findings arrive within the first week.
Adversarial reasoning and 1,200+ attack patterns separate theoretical findings from working attack paths. Each proven path becomes a change your engineers review and merge, with root cause and priority attached.
Every fix is locked as a rule. On each release the rule is re-tested against the live product, so a regression is caught before it ships. Teams that want wider coverage run MATT on developer machines as well, so the rule is checked as the code is written.
We read source code and architecture through GitHub and test with a login you provide. Production data, secrets and customer records stay out of scope.
One audit tells you where you stand. The assurance system re-tests every rule on every release, including code your AI wrote, and gives the board one page on demand.
A fixed-scope assessment across 1,200+ patterns, ranked by business impact, with fixes written alongside your team. You know where you stand within three weeks.
The assurance system keeps every rule under test as the product changes. The answer to "are we still secure?" is current on any given day, not on the day of the last audit.
One page, generated on demand: which rules hold, what changed, what was fixed and when. Maps directly to SOC 2 controls and customer security questionnaires.
A researcher looks at your public deployment live, names the two or three exposures that matter most, and tells you what an audit would scope. You leave with a straight answer.
Continuously discover, validate, fix and prevent what is exploitable in your software.
Your scanners, pentests and evals become one loop that runs on every change. Every discovery becomes a rule that keeps running.
What is ExploitOps?→MATT combines application context, adversarial reasoning and 1,200+ attack patterns to separate theoretical findings from meaningful attack paths. One capability inside Prove.
Ownership checks, session handling, privilege boundaries. Seen at Hugging Face.
Metering, quotas, and the compute your product gives away for free.
Responses, logs, environment variables, third-party calls. Seen at Vercel.
Signups, resets, and transactions at a scale humans never operate at.
What it takes to slow or stop the product. Seen at OpenBSD: two packets.
Generated handlers, agent permissions, prompt-injected tooling. Seen at Black Hat.
Read access through GitHub. Nothing sensitive is touched. Your team approves every change.
A researcher looks at your public deployment live and gives you a first read.
Out: scope and a straight answer
Read-only access. We write the rules your product has to hold, then map what changes across the surface a signup can reach.
Out: your first rule set, approved by engineering
Every rule is run against the live product across 1,200+ attack patterns. Each rule comes back holding or broken, and what broke becomes a fix your team approves.
Out: fixes merged, every rule with a check attached
Every meaningful change observed and every rule re-proven against the deployed product, including AI-written code.
Out: one page for the board, on demand
NDA in 24 hoursRead-only accessMapped to SOC 2 and ISO 27001How procurement works →
A 15-minute call with a MATT co-founder to understand your product and how you handle security today, and whether an ExploitOps Review is worth doing. If there is a real need, we scope one with the research team.