On November 13, 2026, Cohere announced it completed the acquisition of AISLE and folded the Prague-based autonomous vulnerability lab into Cohere Cyber, while also welcoming cubic for AI-powered code review. AISLE had already confirmed on November 10 that the deal closed that morning. Stanislav Fort, who led AISLE research that touched Cursor, VS Code, PyPI litellm wheels, and an MCP server for assistants, becomes Vice President of AI Security at Cohere.
If you run Claude Code, Codex, Cursor, or similar coding agents, this does not ship a new toggle in your IDE. It changes what to expect from a major model vendor: more in-house CVE discovery, enterprise code-review agents, and cyber research wired into training—while your laptop still holds the API keys, MCP configs, and shell access that Beam and your own policy have to watch.
| Question | Short answer |
|---|---|
| Am I affected? | Indirectly if you use Cohere APIs, AISLE/Cohere Cyber products, or depend on vendors they audit (curl, FFmpeg, editors) |
| Is there a fix? | No consumer patch—this is org and product structure; patch your editors and dependencies from vendor advisories as usual |
| Does data leave my machine? | Cohere Cyber sells to enterprises with private or air-gapped options; local coding agents still egress on your network unless you constrain them |
| What is new vs October LOI? | Closed acquisition, Cohere Cyber brand, cubic secure coding vertical, Fort VP role |
| What should I verify? | Editor versions, PyPI/npm pins, MCP allowlists, and whether you assumed “the model vendor handles security” |
Timeline: letter of intent to Cohere Cyber
The deal followed a public arc Cohere and AISLE both documented:
- October 15, 2026 — AISLE announced a letter of intent for Cohere to acquire the company, with Prague slated to become a Cohere office and Fort named future VP of AI Security.
- November 10, 2026 — AISLE posted “Goodbye AISLE, Hello Cohere Cyber”, stating the acquisition formally closed that morning, that engineers had already completed a first cyber sprint with Cohere, and that AISLE reported 650+ CVEs discovered in the prior 14 months.
- November 13, 2026 — Cohere’s blog post announces completion of the AISLE acquisition and cubic joining Cohere, bringing Cohere’s 2026 strategic M&A count to four after Aleph Alpha and Reliant AI.
Leadership per Cohere: Ondrej Vlcek becomes Senior Vice President of Cohere Cyber; Jaya Baloo becomes Cohere CISO; Stanislav Fort becomes VP, AI Security, joining core model research and training. Cohere also plans a long-term investment in Prague as a global center for AI cybersecurity.
What AISLE was building before the buyout
AISLE’s pre-acquisition work matters for coding-agent users because it targeted the same surfaces agents touch: editors, package indexes, MCP, and large C libraries agents compile against.
Editor and assistant attack surface. AISLE’s public research list includes a July 31, 2026 report on a one-click RCE class issue affecting Cursor, VS Code, and Google Antigravity—the kind of flaw that matters when an agent opens untrusted project files or follows malicious deep links. Beam covered related editor trust issues in embedded browser flaws across coding IDEs; AISLE’s findings are upstream vendor disclosure, not automatic protection on your machine.
Package supply chain speed. On October 21, 2026, AISLE reported its dependency firewall flagged litellm 1.107.0 on PyPI with a credential stealer in the wheels roughly three minutes after release—a timeline that matters for agents that run pip install or refresh lockfiles without human review. That sits in the same problem space as malicious npm skills targeting Claude and general MCP supply-chain guidance.
MCP as product. AISLE announced an AISLE MCP Server on November 9, 2026 to expose security findings to Claude Code, Codex, and Cursor via read-only MCP tools—conceptually adjacent to how teams wire Claude VirusTotal vtai MCP or Coral gateway custom tools, but aimed at enterprise AISLE tenants, not local laptop policy.
Library-scale CVE mining. AISLE cited discoveries across FFmpeg, Wireshark, OpenSSL, and curl releases in its LOI post—libraries that show up constantly in agent-generated build scripts and container images. Google GTIG’s exploit-agent patch differential work shows why fast discovery plus slow patching is a recurring agent-risk pattern; AISLE’s pitch was to shorten the discovery side.
Leadership and sovereignty: who moved where
Cohere framed the AISLE deal as cyber resilience inside sovereign AI, not a bolt-on security SKU. The same November 13 post assigns Ondrej Vlcek to run Cohere Cyber as a dedicated business unit, elevates Jaya Baloo to company-wide CISO, and places Stanislav Fort inside core model research and training—a signal that AISLE’s vulnerability work is meant to influence weights and evaluation, not only customer-facing scanners.
That structure matters when you compare vendors. A lab that discovers curl or FFmpeg CVEs at scale can feed training data and refusal policies for models your coding agent calls through an API. It does not automatically harden the Cursor workspace on disk where the agent still runs git, docker, and arbitrary MCP tools.
AISLE’s November 10 closing post also emphasized Prague as a global cybersecurity hub and continuity for AISLE customers under the Cohere Cyber name. If your organization already piloted AISLE, expect branding and account migrations, not an instant MCP entry in every individual developer’s ~/.cursor/mcp.json without enterprise onboarding.
What cubic adds in the same announcement
Cohere’s November 13 post describes cubic as an AI-powered code review platform that reviews changes, finds bugs and vulnerabilities, understands large codebases, and automatically remediates issues, with review agents built for on-premises and air-gapped enterprise environments. Founder Paul Sanglé-Ferrière will lead Cohere’s secure coding vertical.
For coding-agent teams, the overlap is conceptual, not a product swap: cubic-style review agents sit in the CI or enterprise repo, while Claude Code and Codex sit on the developer workstation. A vendor shipping both frontier models and review agents can align safety training with review policy—but it does not remove the need to log tool calls, deny destructive commands, and scan skills before execution on the machine where the agent actually runs.
What this means if you run AI coding agents
Treat Cohere Cyber as upstream research and enterprise services, not endpoint control. Cohere’s sovereign AI narrative emphasizes customer-controlled infrastructure. That is a different layer from local agent hooks that see rm -rf, credential patterns, and MCP JSON on disk.
Concrete checks that still apply after this acquisition:
- Pin and verify packages agents install. AISLE’s litellm example is a reminder to use hash or version locks, private mirrors, and human review on dependency bumps—especially when an agent proposes “just upgrade litellm.”
- Patch editors and extensions on the same cadence as OS security updates. AISLE’s editor research is a reason to track Cursor, VS Code, and JetBrains security advisories, not only language runtimes.
- Inventory MCP servers before enabling them in Claude Code or VS Code. Enterprise cyber MCP from a vendor does not replace allowlisting and heuristic MCP config scans.
- Separate vendor trust from session trust. Cohere previously documented unusual open-weight model egress probes in Qwen3.8 call-home testing; corporate cyber M&A does not retroactively sandbox models you run locally.
- Read MCP docs before pasting enterprise configs. AISLE’s MCP setup examples target
https://your-company.api.aisle.cloud/mcpwith OAuth or PAT headers—exactly the kind of tenant-specific endpoint Beam’s heuristic scan expects you to pin and review before agents store it beside personal keys.
AISLE MCP server: enterprise findings, read-only by design
AISLE’s November 9 MCP announcement documents five read-only tools—issue-list, issue-detail, issue-count, fix-list, and repo-list—so assistants can prioritize proven exploits and explain patches without the server mutating backlog state. The post explicitly positions read-only access as a rollout tradeoff: faster adoption in large orgs, with humans still merging fixes under normal git permissions.
For Claude Code users, the documented pattern is claude mcp add --transport http aisle https://your-company.api.aisle.cloud/mcp with OAuth or a PAT header. That is legitimate security tooling, but it is also another MCP origin to classify: corporate-approved, bearer-tokened, and outside Beam’s default skill heuristics unless you scan the saved JSON. Pair vendor MCP with the same discipline as trending community MCP servers: name the server, document owners, and reject copy-paste configs from unofficial READMEs.
Search questions this news raises: “cohere aisle acquisition” resolves to a closed deal in November 2026, not the October LOI. “cohere cyber” is the AISLE successor brand inside Cohere. “is cohere publicly traded” is outside this incident—Cohere remains a private company in its own announcements; the security angle is product and research integration, not a stock event.
How Beam handles this
Beam overlaps with parts of the Cohere Cyber story—secure software workflows around agents—but not with sovereign enterprise deployment or autonomous CVE mining at library scale.
On a developer laptop, Beam stays local-first: it watches Claude Code and Codex tool use, can deny risky PreToolUse patterns in supported harnesses, scans skills and MCP configs with eleven heuristics plus a version-pin check, and redacts secrets before storage. That is session- and config-level control, not AISLE-style whole-codebase autonomous remediation or Cohere-hosted review agents.
Where Beam is the better fit: teams who want on-device audit evidence and MCP/skill hygiene without routing proprietary repos through a vendor cyber pipeline. Where it is not: replacing enterprise air-gapped Cohere deployments, managed Cohere Cyber services, or AISLE’s CVE discovery lab. Beam’s deny path is hook-based, scanning is heuristic, and missing cached policy is fail-open—limits you should weigh against any vendor’s “AI security platform” claims.
See coding assistant security and what AI agent monitoring tracks.
Frequently asked questions
Did Cohere acquire AISLE?
Yes. Cohere announced completion on November 13, 2026; AISLE confirmed the closing on November 10 and the Cohere Cyber rebrand.
What is Cohere Cyber?
Cohere’s cybersecurity business built around AISLE, focused on autonomous vulnerability discovery and remediation for sensitive deployments.
Who is Stanislav Fort at Cohere now?
Vice President of AI Security, leading AI security research and working with core model training per Cohere.
What did AISLE find in coding tools before the acquisition?
Published 2026 research on major code editors, PyPI litellm wheels, MCP integration, and widespread library CVEs.
Does Cohere Cyber replace local coding-agent security?
No. Local agents still need hooks, scans, egress policy, and dependency discipline on the workstation.
What is cubic in the same Cohere announcement?
An acquired AI code review platform aimed at enterprise and air-gapped secure coding workflows.
How many CVEs did AISLE report discovering?
AISLE cited 650+ CVEs in the 14 months before its November 10 closing post.
Is Cohere AISLE acquisition related to Qwen call-home testing?
No direct link; Qwen egress observations were a separate Cohere research disclosure.
Related reading
- Qwen3.8 call-home behavior in Cohere testing
- MCP security practical guide
- Google GTIG exploit agents and patch gaps
- Phantom053 malware targeting Claude Code users
- What AI agent monitoring tracks
- NVIDIA SkillSpector open-source skill scanner limits
Leadership titles, CVE counts, and acquisition timing follow Cohere’s November 13, 2026 blog post and AISLE’s November 10, 2026 closing announcement as of publication.

