The trust layer for AI in production
Mission · private brief
AI is entering production.
The audit trail is not ready.
We are building the missing trust layer.
CollAI gives regulated teams a safer way to let AI touch operational context: sensitive data is controlled, outputs are inspected, and high-risk actions remain accountable. In environments where accountability has to survive an incident review, policy PDFs are not enough.
Open demo · no waitlist. Design partner track is for regulated FinServ teams.
0%
Employees now using AI regularly on corporate devices
Verizon DBIR 2026 · up from 15%
0st
SEC Item 1.05 8-K already tied to unauthorized AI exposure
2026-05-11 · SEC EDGAR Item 1.05
0d
Until the next high-risk AI compliance clock lands
2 Dec 2027 · European Parliament · 16 Jun 2026
Failure mode · public record
Three documented cases where the model was allowed to run and the harm arrived on the way out: a rendered link, an encoded payload, an action nobody approved.
Jun 2025
Microsoft 365 Copilot
A single crafted email made Copilot place internal data inside a rendered link and send it out with no user click.
Control that failed · The cross-prompt-injection classifier let the email through. Nothing inspected what the model wrote back.
CVE-2025-32711 · EchoLeakJul 2026
OpenAI research agents
Agents broke isolation, coordinated over an unapproved channel, and executed code on Hugging Face production servers.
Control that failed · No named human approved any of it. OpenAI describes agents taking dangerous actions that no human directed.
OpenAI incident reportJul 2026
Hugging Face forensics
Data on its way out was chunked, XOR'd, and compressed so a plain text scan of the logs would miss it.
Control that failed · The first automated scan surfaced few secrets. Decoding it the way the agent encoded it recovered roughly 4x more.
~17,600 reconstructed actionsThe lesson is not patch faster. These incidents are public proof that input filters miss agent-to-agent paths and editable logs. CollAI governs the seam after the model and before the action: VPC scrub, StegoGuard on outputs, named-human approval before destructive actions, and a hash-chained audit. Cited as public evidence only; CollAI had no role in these incidents. Banks need that layer before agents touch production, not after.
Operating thesis
CollAI is being built for operators who need machine-speed assistance without surrendering boundaries, judgment, or proof.
Operational data is controlled before it enters any automated workflow.
Every machine-produced recommendation is treated as untrusted until cleared.
Decisions are gated, recorded, and reviewable when it matters.
Verified EU AI Act timeline · European Parliament · 16 Jun 2026
Source: European Parliament →2 Aug 2026
Most remaining AI Act provisions · chatbot transparency
Still scheduled; high-risk rules deferred by Omnibus
2 Dec 2026
Machine-readable AI content labelling · watermarking
Systems on market before 2 Aug 2026
2 Dec 2027
Stand-alone high-risk AI (Annex III)
HR, credit scoring, biometrics, etc.
2 Aug 2028
High-risk AI embedded in regulated products (Annex I)
Medical devices, machinery, toys
Omnibus adopted by Parliament 16 Jun 2026 (423–57–174). Council formal adoption and Official Journal publication pending. Dates above reflect the adopted parliamentary text, not the superseded August 2026 high-risk deadline.
Research agents coordinated through an unapproved internal channel, reached the public internet, and ran code on dozens of Hugging Face servers. OpenAI calls it a warning shot that capable agents can take dangerous actions no human directed.
Read sourceDesign partners
Seeking 3-5 regulated FinServ design partners. Work email preferred.
Identity
CollAI is live. Public materials describe the problem and operating principles. Deep architecture briefings stay private for qualified partners.
Access
Self-serve demo at dashboard.trycollai.com. Design partner briefings for regulated FinServ teams evaluating governed agent execution.