Enterprises deploying autonomous AI agents face immediate risks regarding data exfiltration, making mandatory security protocols the only viable path forward. On October 6, 2026, the Toronto-based artificial intelligence firm Cohere announced its new lockdown mode feature equipped with strict Access Control Lists (ACLs) to directly address these vulnerabilities.
The company aims to secure North Agent harness operations by restricting how models interact with external tools and sensitive knowledge bases. Pricing details for this security tier remain unconfirmed in official documentation, though supply-chain indicators suggest an entry point around $10,000 per month for qualifying enterprise subscriptions. Prior to this update, organisations utilising agentic workflows encountered significant friction when balancing automation efficiency against corporate data governance. Integration with existing knowledge bases and SaaS applications introduced additional vectors for potential misuse, as agents frequently navigated complex permission hierarchies without adequate safeguards. AI models open.
Cohere announced its new lockdown mode feature:
| Feature | Standard Configuration | Lockdown Mode |
|---|---|---|
| Tool Access | Broad permissions by default | Restricted via fine-grained ACLs |
| Data Permissions | High risk of exfiltration | Explicit role-based restrictions |
| Token Spend | Potential for runaway queries | Optimized endpoint targeting |
| Governance | Manual policy enforcement | Automated structural compliance |
The introduction of strict access controls positions Cohere as a critical player in the battle for secure agentic infrastructure, particularly as competitors face increasing scrutiny over autonomous agent safety. CEO and co-founder Aidan Gomez has consistently emphasized the importance of reliable enterprise deployments, and this release demonstrates a commitment to solving the trust deficit that plagues widespread AI adoption. Organizations evaluating agentic frameworks must now prioritize platforms that offer native lockdown capabilities rather than retrofitting security post-deployment. The timing of this release coincides with heightened awareness regarding autonomous agent risks, as other major technology firms navigate legal and technical hurdles related to model alignment. By offering a structured path to safer deployments, Cohere appeals to risk-averse sectors such as finance, healthcare, and government contracting, where data integrity outweighs experimental flexibility.
Adopt lockdown mode now if data sovereignty matters; skip it only if you accept unlimited agent risk.
FAQs
What is the primary purpose of Cohere’s lockdown mode?
Lockdown mode prevents unauthorized data exfiltration by enforcing strict access controls that restrict how enterprise AI agents interact with external tools and sensitive information. For more detail, see VentureBeat AI.
How does the new feature restrict agent tool usage?
The system utilizes fine-grained access control lists to map specific roles to distinct permissions, ensuring agents can only invoke approved tools and query authorized data repositories.
Is there a minimum cost to access this security feature?
Official pricing remains unconfirmed, but market indicators suggest the enterprise tier subscription starts at approximately $10,000 per month for qualifying organizations.
Where is Cohere headquartered?
Cohere is based in Toronto, Ontario, Canada, under the leadership of CEO and co-founder Aidan Gomez.
Does the update affect existing agent functionalities?
The overhaul retains core constructs like skills, libraries, and automations while wrapping them in stricter security boundaries to maintain workflow continuity without sacrificing safety.
Was this article helpful?
Your feedback directly improves future articles on this site.
