EchoLeak is a warning about the role AI could play in cyber insurance and resilience
Cyber ResilienceArticleAugust 19, 2025
EchoLeak is a challenge to how we assess cyber risk in the AI age
The recent discovery of EchoLeak, a zero-click exploit aimed at Microsoft 365 Copilot, is more than a new technical threat. It’s a wakeup call for how we assess, underwrite and manage cyber risk in an AI-driven world. For cybersecurity, risk, and insurance professionals this is a pivotal moment.
What is EchoLeak and why does it matter?
EchoLeak is a type of attack that manipulates an AI agent, like Copilot, into misinterpreting and acting on content.
A harmful email, which doesn’t need to be opened or clicked, can quietly cause AI to reveal sensitive internal information. The AI reads the email, follows hidden instructions, and unwittingly leaks sensitive data from Outlook, Teams, OneDrive, and SharePoint.
This isn’t a traditional attack. There is no malware or unathorized access. It’s a way of manipulating AI behaviour by taking advantage of how it understands context, and is designed to work, rather than tampering with its code. The implications for cyber risk are profound.
“EchoLeak is a stark reminder that AI-driven systems can be manipulated in ways we’ve never seen before. The threat isn’t just technical, it’s about how we understand, insure, and respond to risk in a world where machines make decisions."
The implications of AI based threats for cyber insurance
Organizations that use cyber insurance as a cornerstone of their cyber risk framework need to reconsider their approach. EchoLeak and similar AI-driven threats raise three immediate concerns:
- Silent cyber exposure
EchoLeak doesn't fit the conventional definition of a cyber event. There's no malware, no unauthorized access, and no system compromise. But the damage is real. It's possible we could see other manipulative events, not just data loss, in the future.
You should speak to your insurer about how your insurance policies will respond.
- AI as a cyber risk multiplier
As AI tools and agents become embedded in our organizations the threat surface expands. When compromised these tools don’t just leak data. They amplify the breach by accessing multiple systems simultaneously.
In some cases, they might access systems across an entire industry. Integration and interconnectedness between systems is designed in. It's why AI is so powerful, but it can also magnify the potential fallout.
- Attribution and claims complexity
EchoLeak muddies the water on attribution, complicating claims handling and subrogation strategies. If an AI agent leaks data then a reasonable case could be made for the responsibility being with the user, the vendor, or the AI developer.
Assessing cyber risks is changing
As AI adoption accelerates, how insurers think about underwriting cyber risk must evolve to consider:
- The scope of AI access: what systems and data are open to AI tools within the organization, and are appropriate risk assessments being conducted and reviewed regularly?
- Prompt injection defenses: are there controls to prevent manipulation and is the organization regularly reviewing the threat landscape when incidents like EchoLeak happen to inform their approach to AI and cyber risk management?
- Vendor transparency: is it possible to audit AI behavior and activity logs? AI is a continually evolving landscape and appropriate safeguards need to be in place such as establishing contractual liability and having a human in the loop for key decisions
- Incident response readiness: is the organization testing its ability to respond effectively to AI-specific scenarios?
- Culture and training: does the organization use its risk view to inform the training for its people who are using AI in their day-to-day work?
- Privileged access management: Where appropriate AI Agents should be treated as privileged users and assessed as a risk in the same way people with enhanced access to systems and data would be
"The real concern is that these attacks can slip past traditional defenses, leaving organizations exposed to new and unfamiliar risks. Staying informed and alert is essential. We’re entering a Zero Trust era, where a healthy dose of skepticism is our strongest line of defense."
What should you do to manage the cyber threat from AI?
Organizations should be reviewing their cyber risk management to ensure:
- Auditing of AI integrations. Know what your AI tools access and how they behave
- Segment trust boundaries: don’t allow AI to blend internal and external content without proper checks and controls
- Implement AI-aware data loss prevention and anomaly detection
- Review insurance policies to ensure AI-induced data loss is clearly addressed
- Push vendors for transparency on how AI handles untrusted inputs
Final thoughts on cyber risk and the new age of AI
EchoLeak is a paradigm shift. It challenges our assumptions about how AI behaves, how risk is transferred, and how insurance coverage is triggered.
In the age of AI, we are not just insuring systems, we are insuring decisions made by machines. The challenge for cyber risk leaders is to bridge the gap between technical complexity and insurable clarity. EchoLeak is our call to action.