Sovereign AI expands beyond model hosting

Cohere has completed its acquisition of AISLE and brought the team behind AI code-review company cubic into the business, expanding its sovereign AI platform into cyber defence and secure software development. AISLE’s autonomous vulnerability discovery and remediation work becomes the foundation of Cohere Cyber, while cubic’s engineers will establish a secure-coding vertical. The move reflects a broader change in enterprise AI competition: customers increasingly buy governed systems that can operate in sensitive environments, not an isolated language model accessed through a public endpoint.

Cohere defines sovereignty as maintaining control of data, infrastructure and operations while still using advanced AI. That promise spans private clouds, on-premises installations and air-gapped environments. Adding security products is strategically coherent because the most regulated customers need to protect both the AI system and the software estate it can access. The challenge is turning several acquired technologies into one auditable platform. A portfolio slide is not the same as shared identity, policy, logging, deployment and lifecycle management.

AISLE specialises in autonomous systems for finding, validating and remediating vulnerabilities. Its leadership is taking senior roles across Cohere Cyber, corporate security and AI security research, and Cohere plans a major investment in Prague as its centre for AI cybersecurity. This puts the acquired expertise close to core model development rather than leaving it as a peripheral product. Customers should still ask which functions are available now, which remain road-map commitments and how remediation proposals are reviewed before they change production software.

AISLE becomes a dedicated cyber business

Autonomous vulnerability work can increase defender reach, particularly across large codebases and legacy systems that receive limited manual attention. It can also create new risk if a model produces an unsafe patch, exposes exploit details or repeatedly probes a fragile environment. A credible service needs isolated testing, reproducible findings, evidence for each vulnerability and a clear approval boundary before remediation. Security teams should measure validated risk reduction and regression rates, not the raw number of issues an agent reports.

The cubic team brings an AI-powered code-review platform designed to understand repositories, identify bugs and vulnerabilities and propose fixes. Cohere highlights support for on-premises and air-gapped deployment, which fits customers that cannot send source code to an external service. Secure coding is an attractive entry point because value can be measured in review time and escaped defects. It also requires careful integration with version control, CI policies and developer identity so an AI reviewer cannot bypass the same controls applied to human contributors.

cubic brings code review into the platform

Cohere says cubic’s review agents perform strongly on code-review benchmarks. Buyers should examine the benchmark tasks, languages and repository sizes, then run an evaluation on their own defect history. Generated findings need severity calibration and deduplication or developers will learn to ignore them. Suggested fixes should include tests and be attributable to a model version, prompt, toolchain and source revision. In regulated development, that provenance may matter as much as whether the patch compiles.

The combination could connect Cohere’s models, Model Vault deployment controls, cyber research and coding agents inside one jurisdictional boundary. That is valuable for governments and critical infrastructure operators that want fewer external data paths. Consolidation also increases supplier concentration. Organisations should retain exportable findings, standard repository workflows and independent security telemetry so they can validate Cohere’s conclusions and change providers. Sovereignty should mean customer control, not merely that one vendor operates more layers of the stack.

Integration quality will decide the strategy

Cohere has completed four strategic transactions this year, including Aleph Alpha and Reliant AI. The pace broadens capability across models, biopharma, cyber security and development tools, but integration risk grows with every new team and product. Customers need a stable support model, consistent release process and clarity about which entity processes their data. Procurement should distinguish company-level announcements from generally available features and require evidence for deployment modes, regional support and security certifications relevant to the proposed workload.

The AISLE and cubic additions make Cohere’s direction clearer: sovereign AI is becoming an operating platform for sensitive work rather than a synonym for private inference. If the components share policy, evidence and deployment controls, the strategy could reduce the friction between creating software and defending it. Australian organisations should test that claim through a contained repository or vulnerability programme before expanding access. A useful pilot should include seeded vulnerabilities, known false positives, rollback exercises and an independent careful expert human review of every proposed remediation. It should also measure developer acceptance, time to validation and whether the combined system can clearly explain disagreements between its coding and security components. The decisive question is whether the platform strengthens accountable human security work while keeping data and change authority under customer control, not whether its agents can generate the largest number of findings.