Cohere Is Buying the Startup That Found 500 Bugs With AI — Security Is Now Part of the Model Sale
A one-year-old startup just became a frontier lab’s security arm
On Thursday, October 15, Canadian AI company Cohere said it had signed a letter of intent to buy AISLE, a San Francisco-based cybersecurity startup that uses AI to find vulnerabilities in software. The financial terms were not disclosed, and a letter of intent is not yet a definitive agreement — this deal is announced, not closed. Even so, the direction is unmistakable, and it says a lot about where enterprise AI competition is moving.
AISLE is barely a year old. It was founded in 2025 by Ondrej Vlcek, the former chief executive of Czech antivirus company Avast, together with co-founders who came from Google DeepMind and Anthropic. That pedigree alone would make it an interesting acquisition target. What makes it a strategic one is the track record the company says it built in its first year: its AI systems found more than 500 previously unknown vulnerabilities in widely used software.
To put the sharpest point on it: AISLE says all 12 vulnerabilities patched in a January security update for OpenSSL were discovered by its systems. OpenSSL is not an obscure library. It is the open-source cryptography software that helps encrypt internet traffic for banks, governments, and messaging services — the kind of code where a missed flaw can have consequences far beyond one company. According to reporting on AISLE’s own account of the deal, its tally runs to more than 500 assigned vulnerability identifiers across more than 100 codebases, including 18 of the 24 fixes in a recent release of curl, the ubiquitous data-transfer tool. The company says those findings were independently verified by the maintainers of the respective projects before being fixed.
A model company is not buying a dashboard. It is buying a demonstrated ability to find real flaws in real, heavily scrutinised code — before someone with worse intentions does.
Why Cohere, why now
Cohere has never positioned itself as a consumer chatbot company. Its bet has always been enterprise and government customers: organisations that care less about a playful assistant and more about deployment control, data residency, and whether a vendor can be trusted inside sensitive systems. That positioning got much bigger in September, when Cohere signed a definitive agreement to merge with Germany’s Aleph Alpha, in a combination reported to be worth about $20 billion. The combined company keeps the Cohere name, operates from Toronto and Berlin, and turns Aleph Alpha’s Heidelberg site into a research centre — pitching itself, in effect, as the first transatlantic sovereign AI provider. Germany’s Schwarz Group, owner of Lidl and cloud provider STACKIT, backed the merger with a major investment and is expected to provide European cloud infrastructure.
Read the AISLE move against that backdrop and the logic snaps into focus:
- Security becomes a product, not a promise. A new division, reported to be called Cohere Cyber, would offer customer-facing security products rather than asking buyers to take safety on faith. AISLE’s Prague engineering operation becomes a Cohere hub, and its leaders are slated for senior security roles in the combined group.
- Private deployment is the common thread. The stated aim is stronger security and threat detection for enterprise and public-sector clients, with private deployment available on the customer’s own systems — exactly the deployment model sovereign and regulated buyers demand.
- Europe is the beachhead. Cohere described the deal as continued investment in Europe. Prague joins Berlin and Heidelberg on the map, giving the combined company engineering depth inside the EU at a moment when European governments are actively looking for AI suppliers they consider governable under their own rules.
In other words, Cohere is assembling the pieces a government procurement officer actually asks about: capable models, infrastructure under regional control, a local footprint — and now, a credible answer to “who checks the code?”
The deeper shift: AI that attacks code, sold by the people who sell AI
For the past two years the cybersecurity conversation around AI has mostly been about risk. More capable models can write exploits, automate phishing, and probe systems at machine speed, and every serious lab now has to explain how it keeps those capabilities from being misused. The AISLE deal highlights the other side of that same coin, and how valuable it has become.
Finding vulnerabilities in mature, widely reviewed open-source code was historically slow, expert work — the domain of a small global community of security researchers doing painstaking manual review, assisted by fuzzers and static analysis. A system that can surface hundreds of genuine, maintainer-confirmed flaws in a year, including a clean sweep of an OpenSSL security update, changes the economics of defence. The bottleneck stops being “can we find the bugs?” and starts being “can maintainers triage, verify, and patch at the speed the findings arrive?”
That is also why this capability is worth acquiring rather than building. A frontier lab could, in principle, point its own models at codebases. What it cannot quickly manufacture is a specialised team — led by a former antivirus CEO, staffed from DeepMind and Anthropic — with a year of verified results against production code and working relationships with the open-source projects that had to confirm and fix the findings. In AI, as in every previous technology wave, the scarce asset is not the demo. It is the proven result.
There is a harder question embedded here, too, and buyers should ask it plainly: a system good enough to find 500 real vulnerabilities is, by definition, a system with serious offensive potential. Whoever owns it takes on the responsibility for how it is used, who gets access, and how findings are disclosed. Cohere’s enterprise and government focus makes that a feature of the pitch — controlled, private deployment for defensive work — but it also raises the standard the company will be held to. Expect customers to ask about disclosure processes, access controls, and audit trails with the same rigour they apply to the models themselves.
What this means if you buy or build AI
- Expect security to be bundled into model deals. The standalone sale — “here is an API, secure it yourself” — is giving way to platforms that arrive with vulnerability scanning, threat detection, and a named security organisation attached. If your vendor cannot describe its security product, that absence will soon count against it.
- Verify claims the AISLE way. The reason this story carries weight is not a benchmark chart; it is confirmed fixes in named, critical projects. When any vendor claims AI-driven security results, ask for the equivalent: independently confirmed findings, patched issues, references you can check.
- Plan for a flood of findings, not a shortage. If AI bug-hunting works at this scale, organisations running large or old codebases should budget for triage and patching capacity, not just scanning tools. Discovery is getting cheaper. Remediation is where the bill lands.
- Sovereignty is becoming a feature checklist. Models plus regional infrastructure plus local engineering plus security is the package now being assembled for governments and regulated industries, on both sides of the Atlantic. Procurement teams elsewhere will start scoring vendors against that checklist whether their own rules require it or not.
The bottom line
A letter of intent is a small piece of paper, and the honest caveats apply: the price is undisclosed, the definitive agreement is still to come, and integrating a specialist security team into a company already mid-merger with Aleph Alpha is real work. Deals like this can still unravel or disappoint.
But the signal does not depend on the closing date. One of the world’s best-funded enterprise AI companies decided that the next thing it needed was not another chatbot feature — it was the team whose AI found every flaw in an OpenSSL security update. Security is no longer a department that reviews the AI strategy after the fact. Increasingly, it is the AI strategy — and as of today, it is something frontier labs are willing to buy outright.