Crypto Exchange Security ⚡ CERT-In AI Warning: Key Theft
CERT-In warning: AI-powered crypto attacks are driving crypto wallet key theft at exchanges. Learn key defenses for crypto exchange security and fintech protection.

AI-Powered Crypto Exchange Attacks Are Moving Faster Than Controls
A warning from India's national cybersecurity agency puts a familiar risk in sharper focus: attackers are using AI tools to move faster against financial companies, including cryptocurrency exchanges. India's Computer Emergency Response Team (CERT-In) says a campaign it calls "CRAFTY FALCON" stole wallet keys and drained funds. It has not named affected exchanges or disclosed losses.
The point here isn't that AI makes attacks fully autonomous. It's that AI can help attackers research targets, generate and refine scripts, and adjust exploits more quickly - while common weaknesses in applications, credentials, and cloud setups still end up doing the heavy lifting. For an exchange, those weaknesses can connect straight to customer assets or systems that move money.
This article breaks down what CERT-In reported, how the attack chain appears to work, and what exchange operators and financial teams can do to reduce exposure. The real question isn't whether a company uses AI. It's whether its security controls can spot and stop a fast-moving intrusion before access turns into withdrawals.
What did CERT-In warn about?
CERT-In says a threat actor it designates "CRAFTY FALCON" is running an active, escalating campaign against financial businesses. Its updated advisory explicitly includes cryptocurrency exchanges, alongside nonbank lenders, digital lending platforms, fintech companies, payment aggregators, and service providers.
The agency warned about the campaign earlier. In its latest advisory, it describes AI-assisted attack tooling and says it continues to receive reports of successful incidents, including from nonbank financial companies and fintech firms that had already received earlier restricted advisories and taken remediation steps. That matters because fixing one issue doesn't necessarily mean every route back into production is closed.
CERT-In reports several outcomes, including stolen cryptocurrency wallet keys and funds drained from wallets. It doesn't name the exchanges involved, quantify losses, or provide enough evidence for the public to connect incidents to specific companies. So the limits of what's known are part of the story: this is an ongoing threat, but not a confirmed list of victims or a total loss figure.
CERT-In characterizes the actor as operating at "machine speed." That language points to pace and scale - not to a system that independently makes every decision. The campaign, according to the advisory, uses AI tools to speed up tasks like generating and refining scripts. Attackers also rely on software weaknesses, exposed test environments, credentials, and access tokens.
A separate report cited alongside coverage of the warning adds context, but it shouldn't be treated as the same CERT-In activity. CrowdStrike reported that an unidentified attacker used ARTEX (an AI penetration-testing tool mentioned in CERT-In's advisory), along with Claude Code and other AI models, in a breach involving South Korean financial firms. The available reporting doesn't prove the actor or campaign matches CRAFTY FALCON.
For exchange operators, the takeaway is operational: treat the advisory as a sign that active risk exists, not as proof that a specific platform has already been compromised. The next step is to verify exposure in your own environment and make sure detection and response work even outside normal business hours.
How does the reported attack chain work?
CERT-In's description follows a common pattern: attackers find a weakness, use it to gain or expand access, and then try to turn that access into a financial outcome. AI may speed up parts of the process, but it doesn't remove the need for a vulnerable entry point or a path to sensitive systems.
1. Find exposed or vulnerable systems
The advisory says attackers used machine intelligence to survey targets and then exploited vulnerable applications and exposed testing environments. A test system can become a serious entry point if it's reachable from the internet, contains production-like data, or shares credentials and permissions with live services.
This risk isn't limited to bugs in public-facing software. An overlooked test deployment, an old service, or a cloud asset with excessive permissions can give an attacker a foothold. In practice, you can't fully protect what you can't reliably inventory, so asset ownership and exposure tracking matter.
2. Reuse credentials and move toward production
CERT-In reports that attackers used shared credentials and access tokens to move from exposed environments into production systems. That transition is where damage accelerates. The initial system may not do much on its own, but if the same secrets or broad permissions connect to systems that process payments, run operations, or manage wallet access, the scope can expand quickly.
That's why patching the first vulnerability often isn't enough if the credentials used during testing are still valid somewhere else. Teams need to know which identities and tokens can reach each system, how long those credentials remain active, and whether access is restricted to the minimum required.
3. Adapt code and scripts to the target
CERT-In says AI tools helped attackers generate and refine scripts quickly. It also reports that attackers reverse-engineered application code to build tailored exploits. In practice, that can reduce the time it takes to adjust an attack method for a specific target. But the underlying reality remains: a weak patch, permissive identity, or exposed service still matters. AI doesn't make a bad foundation irrelevant.
CERT-In also mentions timing around weekends and holidays, when lower staffing can slow review. In some cases, attackers used small test transactions to see whether unauthorized transfers would work before attempting larger movements. The lesson is to treat unusual low-value activity as worth investigation, not to assume only large transfers indicate a breach.
The reported chain ends with clear consequences: stolen wallet keys and drained funds. The advisory doesn't lay out the technical path to every wallet or transfer, so companies shouldn't assume a single wallet setup or transaction method was used. A practical approach is to map and monitor the full route from application and identity access to systems that can initiate or approve value movement.
Why are crypto exchanges exposed to this kind of attack?
Crypto exchanges run internet-facing software, cloud infrastructure, complex internal operations, and systems that can affect digital assets. Each layer has a different purpose and often different owners. The real risk comes from the connections between layers - especially when an access path created for testing or convenience also reaches sensitive production systems.
An exchange may separate customer-facing systems from wallet infrastructure, but that separation only holds if identities, credentials, and network paths enforce it. If an attacker reaches a lower-risk environment and finds a token with broader privileges, the intended boundary may not hold. The same risk exists when teams use shared credentials across development, testing, and production: compromise in one area can spill into others.
CERT-In's description also underlines why "we passed an audit" isn't enough as an ongoing model. An audit looks at controls at a particular point in time. It can't guarantee that a newly exposed test environment, a changed cloud permission, or a compromised access token will be caught before someone uses it. As TRM Labs policy head Ari Redbord has argued, attack techniques can outpace a single review cycle at launch. Continuous review is how you respond to that gap - not a claim that audits don't matter.
There's a similar balance on the defense side. Solana Foundation chief information security officer Michael Coates has said defenders will need access to strong AI tools because attackers will use capable tools too. But adopting AI doesn't automatically translate into better security. Tools still need good telemetry, clear escalation paths, and people who can validate findings and contain incidents.
For an exchange, the business impact can extend beyond stolen funds. A confirmed compromise can disrupt operations, trigger customer and regulatory inquiries, and force difficult decisions about withdrawals, account access, and communications. CERT-In doesn't quantify those impacts for this campaign, but they're practical reasons to build incident response readiness before an incident happens.
So the issue isn't "AI versus no AI." It's whether the organization can see and control the path from exposed application access to production privileges - and ultimately to the systems that move or protect funds. AI can accelerate an attacker through a weak chain. It can't fix a chain that has no reachable weakness or usable privilege.
What should exchanges and financial firms do now?
CERT-In urged at-risk organizations to treat the campaign as active and escalating. It specifically recommended stronger around-the-clock monitoring, increased staffing, and lower alert thresholds during weekends, holidays, and other periods with limited coverage. The logic is straightforward: controls that only work when a full team is online aren't enough for services that operate continuously.
A practical response is to work through the attack path, rather than treating "use AI security" as a complete plan.
Close the easiest routes into production
Start by inventorying internet-facing applications, test environments, and cloud assets. Make sure each environment has an owner, a clear purpose, and a review date. Remove test systems that are no longer needed, restrict access to the ones that remain, and confirm that test setups don't expose production data or share production credentials.
Next, review identities and secrets. Rotate credentials and access tokens that may have been exposed, eliminate shared credentials where possible, and reduce permissions so an account used in one environment can't automatically reach another. A token should have a clear owner, a limited scope, and a defined lifetime.
Monitor the transitions that matter
Monitoring should go beyond login failures or large withdrawals. Look for unusual access between test and production systems, unexpected use of privileged tokens, changes to cloud permissions, new or modified files, and activity that doesn't match normal behavior for payment or disbursement APIs.
For transfers, investigate patterns as well as amounts. A small transaction that tests whether a new route works may matter more than its dollar value. Set thresholds and review rules that account for unusual destinations, timing, authorization paths, and transaction sequences. Then tune those rules with operations teams so alerts are actionable instead of constant noise.
Plan for nights, weekends and holidays
CERT-In highlights low-coverage windows. Identify who reviews alerts during those times, who can suspend sensitive operations, and how incidents get escalated when the primary team is unavailable. Run a real exercise to test the handoff, including steps to preserve evidence and limit further access.
Lowering alert thresholds can help, but it can also create alert fatigue. Make the change targeted: prioritize high-impact systems, wallet-related services, suspicious token use, and unusual disbursement activity. Confirm the on-call team can tell the difference between a real incident and expected maintenance.
Protect the path from access to assets
Map which services and people can initiate, approve, or alter transactions and wallet operations. Review separation of duties, approval requirements, and emergency procedures. If an application account can both change a transaction route and authorize movement of funds, that combination deserves special attention.
Finally, validate fixes after remediation. Re-test the original entry point, invalidate related secrets, review logs for follow-on access, and confirm the containment step didn't leave another usable route. CERT-In's mention of successful attacks at firms that had already undertaken remediation is a reminder to test whether a fix holds across the wider system - not just whether the first visible issue was patched.
Key takeaways
- Treat the campaign as active, but don't assume a particular exchange was affected without confirmation.
- Prioritize exposed test environments, shared credentials, access tokens, and cloud permissions.
- Monitor the route from application access to production services and funds movement.
- Staff and tune monitoring for weekends and holidays, not only standard business hours.
- Recheck remediation and investigate small, unusual transactions as well as large ones.
Conclusion
CERT-In's warning is a reminder that AI-assisted attacks can shrink the time between finding an opening and trying to use it. For crypto exchanges, the risks CERT-In points to are tangible: exposed applications and test environments, reused credentials and tokens, movement into production, and attempts to reach systems connected to funds.
The strongest response isn't a single AI product or a one-time audit. It's a system you keep testing: limited access, clearly separated environments, monitoring that follows sensitive activity, and people ready to respond around the clock. Treat the campaign as active, verify your own exposure, and make sure remediation closes the whole path - not just the first vulnerability you find.