Case study
EchoLeak: a zero-click prompt injection in Copilot
EchoLeak was a critical flaw in Microsoft 365 Copilot in which a crafted email could lead the assistant to leak data with no action from the user. Microsoft said it was fully addressed and customers needed to do nothing.
Last reviewed against NCSC (UK), OWASP Gen AI Security Project and UK Government (DSIT). Sources
The facts
What was reported
Researchers at Aim Security found the flaw, tracked as CVE-2025-32711, and described it as the first known zero-click attack on an AI agent [7]. A remote attacker could steal sensitive organisational data by sending an email, with no action from the target [7]. The reporting calls the underlying problem an "LLM scope violation", where untrusted outside input can direct the model to access and leak privileged data [7].
Microsoft's response
Fixed, no customer action
Microsoft said it had coordinated with the researchers for months and that its advisory said the issue was "fully addressed and no further action was necessary by customers" [7]. It said it was adding defence-in-depth measures [7]. The researchers said there was no evidence that any customers were targeted [7]. Treat that as the researchers' statement, not an audit of every tenant.
The mechanics
Defences that were bypassed
The point for buyers is that a prompt injection classifier was one layer, and layered controls were what mattered [8].
A later academic case study analysed how EchoLeak chained several bypasses to cross trust boundaries without user interaction [8]. It lists these defences as evaded:
- Microsoft's Cross Prompt Injection Attempt classifier
- Link redaction, circumvented with reference-style Markdown
- Auto-fetched images
- A content security policy, through an allowed Microsoft Teams proxy
Lessons
What it teaches
Scope matters
The assistant could reach private data and was reading outside content, which is the pairing the lethal trifecta warns about [6].
Filters are not enough
The case study argues for least privilege, defence in depth and continuous adversarial testing [8].
Vendors can fix single flaws
A fix to one flaw does not remove the wider weakness, so the NCSC's advice to design for residual risk still applies [1].
Do you need to act on EchoLeak itself?
Microsoft said customers did not need to [7]. The useful action is to review the access any assistant has to your mail and files, using the exposure checklist.
Keep reading
Read next
AI browsers
OpenAI says prompt injection in AI browsers is unlikely to be fully solved. What that means for UK organisations that let agents browse and act.
Read the guide
Ask suppliers
A buyer's list of questions on prompt injection for AI suppliers, drawn from NCSC and OWASP advice, for UK procurement and security leads.
Read the guide
Monitoring
What to log and watch for when AI tools may be attacked through prompt injection, based on NCSC and OWASP advice for UK security teams.
Read the guideHow CyPro can help
Need a hand with any of this?
Patching is one job. Keeping a vulnerable system safe until it is patched, and proving it, is several more. CyPro's consultants, testers and SOC analysts can take any of them off your plate.
Vulnerability scanning
Find every internet-facing system you run, including the forgotten ones, and see which are still on a vulnerable build.
See the scanning service
Hardening and segmentation
Our consultants put the controls in place that limit the damage when something is missed: firewall rules, network isolation and tighter admin access.
More on cypro.co.uk
24/7 security monitoring
Our SOC analysts watch your systems day and night for the signs of an attack in progress: unexpected processes, new outbound connections, logins that do not fit.
More on cypro.co.uk
Penetration testing
A focused test of your systems and the controls around them, so you know they hold before an attacker finds out for you.
More on cypro.co.uk
Incident response
If you may already be compromised, our responders contain it, preserve the evidence, find out what was touched and get you running again safely.
More on cypro.co.uk
Fractional CISO for the board
A senior security lead who advises the board on what to fix first, what to accept and what to tell customers, and owns the decision record.
More on cypro.co.uk
Know your exposure
Not sure you are protected?
Prompt injection cannot be patched away, so the useful question is how far a manipulated tool could reach. Testing your AI-connected systems the way an attacker would shows where the limits on data and actions need tightening.