When AI remembers the wrong thing
Saved information can make AI more useful. Leaders need to decide when a remembered claim is allowed to influence a business decision.
Read the perspectiveAI security perspectives · By Henry Awere
Securing Intelligence
AI security decisions for business leaders.
Our flagship thought-leadership series explores how AI changes business risk—and the decisions that protect customers, operations and trust. Written for business owners, executives, and technology and risk leaders.
The business conversation
Perspectives for leaders
Each article takes one question, explains why it matters and offers a clear point of view on what to do next.
Securing AI Agents: Understanding the Architecture and the Risks
As AI agents gain greater access to business systems and the ability to execute actions, organizations need a clearer understanding of how their design influences security risk.
When an AI upgrade changes your business risk
A useful new feature can give an assistant more authority. Make that change a deliberate business decision.
Who can stop your AI when something goes wrong?
The ability to intervene should be part of the adoption decision. Leaders need to know who can act, what will stop and how the business will keep operating.
A good AI answer can still hide a business risk
When an assistant can use tools, the result is only part of the story. Leaders also need confidence in how the work was carried out.
Put the ideas into practice
Companion guides
For technology, security and risk teams carrying out the work: detailed methods, worked tests and evidence to support the decision.
Securing persistent memory in AI agents
Follow the supplier example through a two-session test, interpret the possible results and use the memory review checklist.
Open the guideReviewing AI connector changes
Use the configuration comparison, four worked test results and downloadable change-review workbook with your technical reviewer.
Open the guideStopping an AI agent safely
Follow ten queued jobs through a worked containment exercise, with evidence criteria, a restart example and an exercise record.
Open the guideHow to assess an AI agent’s actions
Use the detailed action-review method and editable seven-field checklist to connect tasks, actions, permissions, approvals and results.
Open the guideEditable resources
Checklists and workbooks
Use the companion guide to understand the method, then record your own scope, observations and findings.
AI connector change review
A blank review form, a seven-field test record and a separate hypothetical worked example.
Download the workbookAI agent action review
Connect the requested task, actual action, target, permission, approval, execution result and finding.
Download the checklistFrom the newsletter
Selected editions
Explore these published perspectives from the Securing Intelligence newsletter on LinkedIn.
Why Runtime Security Is Non-Negotiable for AI Agents
A perspective on agent authority, tool access, policy enforcement and containment during live operation.
Read on LinkedInFrom Pilot to Production: A Systematic Risk Assessment Framework for AI Agents
A framework for examining agent capabilities, business consequences and layers of control before deployment.
Read on LinkedInWhy AI Governance Only Scales When It Lives in the Workflow
An examination of how governance expectations connect to the way AI systems are built and deployed.
Read on LinkedInAbout the author
Henry Awere
Henry brings experience across AI governance, cybersecurity, privacy and technology risk. Through Securing Intelligence, he examines the business decisions behind secure and accountable AI adoption.
Strategic Outlook Advisory supports small and medium-sized organizations and public-sector leaders in Canada and international markets.