Cohere Buys Its Way Into Security: AISLE and cubic Close Out Four Deals in a Year
On Nov 13, 2026, Cohere completed its acquisition of AISLE and welcomed the cubic team, closing out four strategic deals in 2026. AISLE becomes Cohere Cyber; cubic builds the secure coding vertical. No prices were disclosed — and no product timeline either.

In short: Cohere filled in two pieces of the puzzle in a single day
On November 13, 2026, Cohere announced two things: the completion of its acquisition of AISLE, and the arrival of the team behind cubic, an AI code review platform. AISLE builds autonomous systems for finding and fixing software vulnerabilities; cubic builds AI code review. Announced on the same day, both point in one direction: Cohere wants to grow from "a company that makes models" into "a company that sells a whole sovereign AI platform to enterprises and governments."
The announcement is short, but it carries weight. It is not just two deals — it closes out Cohere's year-long strategy. With these two, Cohere has completed four strategic M&A transactions in 2026, after Aleph Alpha and Reliant AI, spanning frontier AI, biopharma and healthcare, cybersecurity, and software development. Four acquisitions in a year is an aggressive pace for a model company.
Who is AISLE: a company that helps defenders find vulnerabilities
AISLE builds autonomous systems for vulnerability discovery and remediation. The idea is to scan codebases with specialized AI systems, find vulnerabilities, verify that they are actually exploitable, and propose fixes — as automated as possible, so defenders move faster than attackers. Cohere's blog puts it this way: such systems can "identify, validate, and remediate vulnerabilities at a speed and scale that helps defenders stay ahead of an increasingly complex threat landscape."
AISLE has published some numbers of its own: in its first year, its platform found more than 500 previously unknown vulnerabilities across more than 100 codebases, each later assigned a CVE number; by November 10, 2026, when the acquisition formally closed, that figure had grown to more than 650 CVE-listed vulnerabilities found over fourteen months. Note that these are AISLE's own figures — Cohere's blog does not repeat them — so they should be cited as company claims, not independently verified results.
The deal had been telegraphed for weeks. Cohere announced its intent to acquire AISLE on October 15, 2026, saying both sides shared a belief: the future of AI depends on securing both AI systems and the software infrastructure behind them. On the morning of November 10, AISLE announced the acquisition had formally closed and that it would serve customers as Cohere Cyber from that day; on November 13, Cohere's official blog wrapped both stories into one announcement.
The people moves were spelled out in full. All three AISLE co-founders take senior roles at Cohere: Ondrej Vlcek becomes Senior Vice President of Cohere Cyber, leading the new cybersecurity business; Jaya Baloo becomes Cohere's Chief Information Security Officer, overseeing security across the company; Stanislav Fort becomes Vice President of AI Security, joining core model research and training while leading AI security research. The founding team staying on, rather than cashing out and leaving, matters a great deal for whether an acquisition like this works.
Cohere also announced a substantial long-term investment in Prague, expanding AISLE's existing presence and making the city its global center for AI cybersecurity and autonomous security systems. That is both reassurance for the AISLE team and a signal to European customers: the security business is anchored in Europe, closer to the clients most sensitive about data sovereignty.
Who is cubic: AI for code review
cubic, on the other side, is an AI-powered code review platform. Its product helps engineering teams review code changes, find bugs and vulnerabilities, understand complex codebases, and automatically remediate some issues. Cohere says cubic's review agents have "consistently ranked among the strongest performers on code review benchmarks" and were built from the start for enterprise environments, including on-premises and air-gapped deployments.
The same caution applies here: "ranked among the strongest" is Cohere's own wording, with no specific benchmark or ranking given. It can be quoted, but must be labeled as a vendor claim, not an independent test result. That kind of phrasing is standard in acquisition announcements, and readers should be able to tell facts apart from marketing language.
The cubic team, led by founder and CEO Paul Sanglé-Ferrière, will establish Cohere's secure coding vertical, integrating enterprise-grade coding capabilities into the platform while meeting customer requirements for security, governance, and deployment. One wording detail is worth recording accurately: Cohere says it "completed the acquisition" of AISLE but "welcomed the team" behind cubic. The two transactions are not structurally identical, and the announcement discloses no financial terms for either. That distinction should be reported as it is, not smoothed over into "acquired two companies."
What the four deals add up to: "sovereign AI"
To understand these two deals, you need Cohere's favorite phrase: sovereign AI. The idea is that enterprises and governments deploying AI should not have to choose between cutting-edge models and control. They should get both: advanced models, plus ownership of their data, infrastructure, and operations.
That positioning is not new, but the demand behind it has sharpened over the past year. As enterprises and governments push AI into workflows that actually matter — cybersecurity, software development, mission-critical operations, decision-making — their requirements have become clear: AI must be powerful, secure, and under their control. It must run in private clouds, on-premises, even air-gapped networks, and meet strict privacy, governance, and compliance requirements. Cohere's blog spends much of its length on this point, because it is exactly where the company tries to differentiate from general-purpose model vendors.
The four acquisitions follow that logic piece by piece: Aleph Alpha adds frontier models and European presence, Reliant AI adds biopharma and healthcare, AISLE adds cybersecurity, cubic adds secure coding. None of them is a standalone product bet; each is a piece that makes the phrase "sovereign AI platform" more credible. In Cohere's words, bringing together frontier models, secure deployment infrastructure, cyber defense, and purpose-built applications builds a platform that gives organizations "more capability, more security, and more control."
What this means for vibe coders
An honest statement first: these deals target enterprise and government customers. No independent developer will find a new "Cohere Cyber" button in Cohere's product list tomorrow, nor a free code review tool next week. Do not expect this news to change your workflow overnight.
But the indirect effects are real and worth watching. First, AI code review is moving from a "nice extra" to a formal line item on enterprise procurement lists. A company like cubic being absorbed by a bigger vendor signals that the market has decided: AI that only writes code is not enough; AI that reviews code, finds vulnerabilities, and fixes them has commercial value. If you build projects for clients, a deliverable that includes an "AI review report" may soon shift from a bonus to an expectation.
Second, security is shifting left, toward the start of the development process. The old pattern was a security check just before launch; the AISLE-style approach finds and verifies vulnerabilities before the code is even merged. For vibe coders, that translates into a concrete habit worth building: ask the agent for an explicit "security self-check" in your prompts, and run a vulnerability scan after code is generated and before it is committed. That habit costs almost nothing today, but it prevents a whole class of problems that only surface after launch.
Third, the sovereign deployment mindset has lessons for independent developers too. You may not need an air-gapped network, but the principle — data and control stay in your hands — applies to everyone: when choosing a tool, ask "will my code be used for training?", "can it run locally?", "can I export everything at any time?" The answers to those questions determine whether a tool is usable long-term more than any feature-list highlight.
Four things the announcement does not say
First, the money. No financial terms were disclosed for either transaction. Without knowing what Cohere paid, there is no way to judge how much premium it paid for these businesses, or how long it can keep acquiring at this pace. Cohere also announced in October that it expects to invest more than 4 billion Canadian dollars in R&D and AI infrastructure in Canada between 2024 and 2031. Acquisitions plus infrastructure is a big footprint; where the money comes from and how long it lasts is a fair question.
Second, a product integration timeline. The announcement does not say when Cohere Cyber will ship specific products, or when cubic's code review capabilities will appear inside Cohere's platform. There are usually several quarters between a closed deal and a shipped product. In the meantime, whether AISLE's and cubic's existing customers keep getting normal service is a question the announcement does not answer.
Third, that 650 CVE number. Finding vulnerabilities in volume is impressive, but the more useful questions are: how severe were they? What fraction got fixed? How many were found in real customer environments rather than public codebases? The announcement does not go there, and readers should not judge by the size of the number alone.
Fourth, the competitive picture. AI code review is a crowded field — GitHub, GitLab, and every major model vendor are pushing in the same direction. Cohere's bet is "security plus sovereign deployment": competing not on whose model is smartest, but on who can get into the most tightly regulated environments. That strategy only works if enough customers are willing to pay a premium for control. That assumption has not yet been validated at scale.
Competitors are pushing the same way
Cohere is not the only company moving from "selling models" to "selling whole systems." In the second half of this year, similar moves have come one after another. Meta acquired multi-agent coding tool Conductor, then announced its cloud service would shut down — a clear sign the technology was being pulled inside Meta's own stack rather than kept as an independent product. Anthropic opened private beta for self-hosted Claude Managed Agents, letting agents run in the customer's own data center with model calls going outside — nearly the same direction as Cohere's sovereign deployment, by a different route. OpenAI launched its Daybreak security product line in the same week, for red-teaming and vulnerability fixing.
Taken together, these moves point to a new consensus: model capability gaps are narrowing, and the real dividing line is becoming "who can hand AI to customers safely and controllably." Teams used to compete on benchmark scores; now they compete on deployment options, compliance, security reputation, and service depth. Scores still matter, but they are no longer the only card to play.
For Cohere, that consensus is both opportunity and pressure. The opportunity is that its sovereign positioning rides the trend — the more regulated the industry, the more it needs exactly what Cohere is selling. The pressure is that rivals are patching the same gap, with larger scale and deeper pockets. Anthropic, OpenAI, or Meta getting serious about sovereign deployment would squeeze Cohere's differentiation fast. How long the window bought by four acquisitions lasts depends on whether Cohere can turn acquired technology into shippable products before competitors catch up.
One more variable deserves attention: open-source models. Cohere's sovereign story rests on "our platform, our models, our deployment." But more and more enterprises are considering a different route: open-source models on their own infrastructure, with maximum control and often lower cost. If open models keep improving, the appeal of Cohere's semi-open route gets squeezed. The announcement does not mention open-source models — unsurprising, since no vendor advertises for its rivals in its own acquisition post — but readers should keep that thread in mind.
Three things independent developers can do now
First, add an "AI review" gate to your own projects. There is no need to wait for Cohere's products to ship, and no need to buy enterprise tools. What you can do today: before committing code, have the AI review its own code. In your prompt, ask explicitly for three categories of problems — security issues (SQL injection, XSS, hard-coded secrets), logic errors (edge cases, null handling, concurrency), and maintainability (naming, duplication, missing comments on critical logic). Then review the flagged issues yourself and decide what to fix. Run that loop for a month and you will have a realistic sense of what AI review catches and what it misses.
Second, build the "shift security left" habit. The core idea behind AISLE and cubic is moving security checks from just-before-launch to while-code-is-being-written. The cheapest version for a vibe coder: after the agent writes code that touches user data, payments, or authentication, follow up with one question — "does this code have security issues, and where would an attacker start?" That follow-up costs nothing but catches many cases where "the feature works but the security has holes." A step further: write common security checks into your project's development guidelines, so every generated piece of code passes them by default.
Third, re-read the data terms of the tools you use. "Sovereign AI" sounds far from independent developers, but its core question — who owns my code and data, and what will they be used for — is close to everyone. Spend half an hour reading the privacy terms of the two or three AI coding tools you use most, looking for three sentences: will code be used for training, how long is data kept, and can you demand deletion? If the answers bother you, it is worth switching tools, or at least moving your most sensitive projects to local models. That half hour is worth more than following any acquisition headline.
Seen in this year's bigger picture
One clear change in the AI industry in 2026: companies no longer compete only on whose model is bigger and smarter. They compete on who can deliver AI into customers' hands and let them use it with confidence. Meta buying Conductor, Anthropic pushing self-hosted Managed Agents, OpenAI launching security products, Cohere making four acquisitions in a year — all of these point the same way: from "selling models" to "selling complete systems that actually land."
For vibe coders, the practical meaning is that the tools you use will increasingly carry "enterprise-grade" attributes — permission management, audit logs, private deployment, security scanning. Features that used to belong to big companies will gradually sink down into products independent developers can use. Understanding the logic behind them is worth more long-term than learning the buttons of any single tool.
Still, keep some skepticism. An acquisition announcement is, at heart, a story a vendor tells the market, and every sentence in it has been carefully chosen. Whether AISLE and cubic are really as strong as the blog claims, and whether Cohere's sovereign AI platform can truly come together, will be decided by product moves over the next few quarters, not by this one blog post. The right way to follow this story is to note the direction, watch for follow-up product launches, and avoid rushing to conclusions.

Summary
On November 13, 2026, Cohere announced the completed acquisition of AISLE and the arrival of the cubic team, bringing its 2026 strategic deals to four. AISLE becomes Cohere Cyber, its three co-founders taking charge of the cybersecurity business, company-wide security, and AI security research, with Prague as the global AI cybersecurity hub; the cubic team will build the secure coding vertical. No financial terms were disclosed for either deal, and no product integration timeline was given. For vibe coders, there is no new tool to use tomorrow — but the direction is worth remembering: AI code review and shifting security left are moving from bonus to baseline, and when choosing tools, asking who owns your data and control matters more than one more feature highlight.
Comments (0)
Sources
Related articles
On 18 November 2026, Google Research and co-authors posted BuggyMLE to arXiv: bugs injected into 222 working ML repositories, and 14 models slide from 95% on the easiest tier to 9% on the hardest. A 9-billion-parameter model trained easy-to-hard beats Claude Opus 5.5 on the 10-bug tier.
On 20 November 2026, OpenAI released MedChemBench: 99 medicinal-chemistry problems from 117 papers and patents, covering 761 molecules. GPT-6.2 Astra scores 84.5% overall — but 95.4% on assembling structures and only 50.4% on choosing compounds. The total flatters; the components tell the truth.
On 20 November 2026, Thinking Machines Lab released Inkling 2 Mini: a 26.4B-parameter looped transformer that runs its middle 12 layers three times per token, fits in about 20GB, and runs offline on a laptop. It leads open-weights models of its size on DeepSWE and SWE-bench Verified — with the caveats printed in the same announcement.