If you wanted to break into the FBI, you would not start at the FBI. You would start with one of the companies the FBI trusts.
In the last few weeks, that is what happened. The ShinyHunters group breached FBIJobs.gov, the Bureau's jobs portal, not by defeating the FBI's own defences but through a side door: a platform managed on the FBI's behalf by a third-party contractor. According to Reuters, citing two unnamed sources, the platform was Oracle PeopleSoft and the contractor worked for Accenture. Oracle had issued fixes for the PeopleSoft weakness the group was exploiting in June, and the FBI blamed the contractor's missed patch.
The FBI acknowledged the unauthorized activity and, per Reuters, removed the contractor. But the sequence matters more than the names: one of the most security-conscious organisations on earth was reached because a platform in its estate missed a patch for months, and nobody's process caught it.
If the FBI's vendor chain can carry a known-vulnerable platform all summer unnoticed, what is sitting unpatched inside yours?
What actually failed
Nobody stormed a firewall here. What failed was a management model: the way nearly every organisation handles third-party risk.
It works like this. Before a vendor is approved, it fills in a questionnaire. Once a year, someone reviews its certifications, policies and attestations. A box gets ticked: vendor reviewed. Then everyone goes back to work, while the vendor's platform changes every day: patches applied or not, configurations drifting, AI features quietly switched on.
The problem was not a missing assessment. Assessment, as practised, is a photograph: it certifies a moment. A platform can be vulnerable from June onward and an honestly answered questionnaire can still miss it, because periodic review was designed to document the estate, not to watch it.
Your estate is no different. Your security is now the security of every vendor, platform, integration, AI tool and contractor-managed system plugged into it: each a door you do not control, all changing daily. Attackers know this. They do not attack the strongest wall; they shop the vendor list.
Why traditional security keeps missing it
Traditional security firms were built for a static world: an annual penetration test, a compliance checklist, a vendor review cycle measured in quarters. That cadence made sense when estates changed slowly.
They no longer do, and the newest layer changes fastest of all. Companies adopt AI agents and tools faster than their inventories can track. A patch missed in June is a breach headline in the autumn. A control checked once a year is, for the other 364 days, an assumption.
The gap is not effort or intent. It is architecture. You cannot close a continuous problem with a periodic process.
How Redthread fights it: assess the estate continuously, not annually
Redthread builds two products around a single conviction: trust has to be verified continuously, and findings have to be proven, not inferred.
The first is our AI-native application protection platform. It is agentless-first, so it installs nothing and starts in minutes. It assesses the estate continuously, including the layer traditional reviews handle worst: the third-party and AI surface. Shadow-AI discovery finds the tools, agents and integrations your teams actually use, including the ones nobody inventoried, which are precisely the ones carrying your risk. Cloud posture and data-security scanning cover the infrastructure underneath. And where it reports a weakness, it proves it: exploit validation reproduces the attack instead of assigning an inferred severity score from a scanner's guesswork.
That changes the vendor conversation. A missed patch cannot hide inside a questionnaire answer: under continuous re-assessment it surfaces in days, as a proven finding with reproduction steps, not at next year's review. All of it sits on one context graph of the estate (applications, code, cloud accounts, identities, the controls already running, and the data they can reach), so a vendor's weakness is ranked by what it can actually reach in your environment. And when a vendor meets the bar, Redthread issues a Letter of Attestation mapped to the frameworks auditors actually ask about: SOC 2, ISO 27001, ISO 42001, PCI DSS v4.0, GDPR, HIPAA, NIST CSF, NIST AI RMF and the EU AI Act. That is evidence a vendor can be held to, not a self-declared checkbox, and firms can issue those reports under their own name as part of how they hold their own supply chain to account.
The same idea, applied to the agent supply chain: ATPP
The second product, our Agentic Trust & Protection Platform (ATPP), applies the same model to the fastest-changing vendor surface: AI agents and the tools they call.
Every day we scan the entire public MCP registry and sign the results, with the day's manifest anchored in the public Sigstore Rekor transparency log and tool definitions fingerprinted so silent changes show. Our 2 November scan covered 55,375 servers in the official MCP registry and found 1,518 listing swaps, where the code behind a listing changed hands under the same name, and 2,193 tools whose definitions changed without a version bump. That is the FBIJobs failure replayed at ecosystem scale: the thing you approved is silently no longer the thing you are running. Publishers whose servers pass earn a public badge, and anyone can check a tool's trust state before depending on it.
Inside an organisation, ATPP gives every agent a cryptographic identity bound to the build actually running, and verifies identity and authority at call time: when the agent acts, not when it was onboarded. Signed trust-state events flow continuously, so if a tool drifts or a proof fails, trust degrades or is revoked as a live signal. For teams ready to go further, the Enforce gateway, in preview, can act on that signal inline.
Build it with us
We are a young company with production systems and a specific point of view: the winners of the next decade of security will be those that stopped certifying trust annually and started verifying it continuously.
We are opening two to three founding design-partner slots. The pilot is free for 30 to 45 days, run on one live part of your estate, vendor and AI surface included. Founding partners get deeply discounted pricing, locked, and a direct line into what we build next. If you are a founding team shipping agents into enterprise reviews, or a company whose vendor list has outgrown its ability to watch it, we should talk.
Start with the truth: run the free self-serve scan, or see the live trust data at trust.redthreadsec.com. The only question is whether you find the open door first, or read about it later.