You can audit smart contracts. You can’t audit yourself. And AI just made human manipulation infinitely more convincing.
The Security Problem Nobody Wants To Admit
The crypto industry has spent billions on smart contract audits, multi-signature wallets, hardware security modules, penetration testing, and bug bounties.
All of it assumes the attack vector is technical.
It’s not.
The Solana Foundation’s new CISO Michael Coates said it publicly this week: crypto’s biggest security threats in 2026 are increasingly coming from AI-powered social engineering and compromised credentials. Not smart contract exploits. Not protocol vulnerabilities.
People.
The attackers shifted targets. They’re not trying to break the code anymore. They’re trying to break you.
And AI just gave them tools to do it better than ever.
What Social Engineering Actually Means
Social engineering is the art of manipulating humans into doing things that compromise security.
It’s not new. Con artists have always existed. Phishing emails have been around for decades. Fake customer support calls are as old as telephones.
But here’s what changed in 2026:
AI made social engineering indistinguishable from reality.
Before AI: A phishing email had grammatical errors, strange formatting, a slightly off email address. Trained eyes could catch it.
After AI: A phishing email is grammatically perfect, emotionally calibrated to your specific psychology, sent from a domain that looks exactly right, at a time when you’re most likely to be distracted, referencing real details from your public profiles.
Before AI: A fake customer support call had an accent, a script, tell-tale signs of inauthenticity.
After AI: A deepfake voice replicates your exchange’s actual support team. The conversation flows naturally. It knows your account details because it scraped your public information. It knows how to build rapport before asking for anything.
Before AI: A fake emergency message from a colleague was detectable because it didn’t sound like them.
After AI: It sounds exactly like them because AI trained on their communication style, their LinkedIn posts, and their email patterns.
The human brain evolved to detect threats from other humans. It didn’t evolve to detect threats from AI systems trained specifically to exploit human psychology.
Why Crypto Is The Perfect Target
Every industry faces social engineering. But crypto has properties that make it uniquely vulnerable.
Irreversibility. When someone tricks a bank customer into a wire transfer, there’s a chance, small but real, of reversal. When someone tricks a crypto user into sending funds, it’s gone—permanently. No chargeback. No fraud department. No appeal.
Pseudonymity. Attackers are harder to trace. The accountability that discourages fraud in traditional finance is weaker in crypto.
High Stakes In Individual Wallets. A single compromised wallet can contain life-changing sums. The ROI on targeting a crypto user versus a traditional bank customer is significantly higher.
Community Of Sophisticated Users Who Think They’re Immune. This is the most dangerous property. Crypto users tend to be technically sophisticated. They know about phishing. They know about scams. They think they’re too smart to fall for it.
That confidence is the vulnerability.
The most effective social engineering targets people who think they can’t be manipulated because they’ve stopped being vigilant.
The Attack Pattern That’s Working Right Now
Coates described the shift clearly: attackers are targeting people, not protocols.
Here’s what that looks like in practice in 2026:
The Fake Emergency: You receive a message, voice, text, or email that appears to be from your exchange’s security team. There’s been suspicious activity on your account. You need to verify immediately or face suspension. The urgency is real. The consequences feel immediate. You act without thinking carefully.
The message was AI-generated. The voice was deepfaked. The urgency was engineered.
The Compromised Colleague: Someone in your organization receives what appears to be a message from a trusted colleague—perhaps your CFO, your CTO, your CEO—asking for a wallet transfer. The tone is right. The context makes sense. The request is urgent because there’s a deal closing.
The colleague never sent it. Their communication style was scraped and replicated.
The Too-Good-To-Be-True Opportunity: You’re approached on LinkedIn, Discord, or Telegram by someone who seems genuinely informed about your project, your portfolio, your interests. They have an opportunity—an early investment, an exclusive access, a partnership. The conversation feels real over days or weeks.
It’s AI maintaining a relationship at scale, designed to eventually extract something.
The Recovery Scam: You posted publicly about a crypto problem. Someone, AI or AI-assisted, found it immediately and reached out offering help. They’re helpful, knowledgeable, and patient. They walk you through “recovery steps” that actually compromise your wallet.
All of these work on smart people. Because intelligence doesn’t protect against emotional manipulation. It often makes it worse—smart people are better at rationalizing why the exception is real this time.
The Quantum Problem In The Background
While social engineering is the immediate threat, Coates also flagged what’s coming: quantum computing.
Post-quantum cryptography is no longer a theoretical concern. Anthropic’s AI recently broke a post-quantum cryptography candidate, raising serious questions about the security assumptions underlying current encryption.
Solana is evaluating post-quantum cryptography. Other chains are doing the same.
This is a technical problem that technical solutions can address. Unlike social engineering, which targets humans, quantum threats target mathematics. Mathematics can be upgraded.
But here’s the uncomfortable overlap: the transition to post-quantum cryptography will itself become a social engineering attack surface.
Users will receive communications claiming they need to “upgrade their wallet security” or “migrate their funds to quantum-resistant addresses.” Some of those communications will be legitimate. Some will be AI-generated attacks designed to look legitimate during the transition.
The technical threat and the human threat converge.
Why “Just Be Careful” Isn’t A Solution
The standard advice: be careful. Verify before you act. Don’t click suspicious links. Check email addresses carefully. Never share your seed phrase.
This advice was adequate when social engineering was low-fi, when attacks were detectable by someone paying attention.
It’s not adequate anymore.
Coates said something important: crypto must “meet users where they are” instead of expecting them to act as security experts.
That’s an acknowledgment that the current model—educate users, hope they stay vigilant—is failing.
Because AI-powered social engineering doesn’t require users to make obvious mistakes. It requires them to make very small lapses in judgment at carefully engineered moments.
You’ve been careful a thousand times. The attack only needs to work once.
What Actually Protects You
If human vigilance is insufficient, what works?
Systems That Don’t Require Perfect Human Judgment.
Multi-signature requirements that mean no single person can authorize a large transfer alone. Time delays on large transactions that create a window for human review. Anomaly detection that flags behavior inconsistent with your patterns.
These aren’t exciting. They’re friction. But friction is the point.
The best security doesn’t make you smarter. It makes the attack harder even when you’re not being smart.
Verification Protocols That Don’t Rely on Communication Channels.
If a “colleague” sends an urgent transfer request, the verification doesn’t happen over the same channel. It happens via a pre-established out-of-band protocol—a specific phone number, an in-person confirmation, a code word.
AI can replicate communication channels. It can’t replicate physical presence or pre-established secrets.
Institutional Humility.
The most dangerous users are the ones who’ve never been fooled because they believe they never will be. The most secure users are the ones who assume they’re vulnerable and design their behavior accordingly.
Security isn’t about being smarter than the attacker. It’s about designing systems that work even when you’re not at your best.
The Industry’s Uncomfortable Admission
Coates’ statement represents something significant: a major blockchain foundation publicly admitting that the threat model has shifted.
For years, the crypto security conversation was dominated by smart contract audits, protocol security, code review. The implicit assumption: the humans are fine, the code needs protecting.
Now the CISO of a major blockchain foundation is saying: the humans are the vulnerability. The code is (relatively) fine.
That’s a meaningful shift, and it has implications for how the entire industry thinks about security.
You can’t audit your way out of this one. You can’t write a bug bounty for human psychology. You can’t patch the vulnerability that makes people respond to urgency.
The security stack has to include the human layer, not just user education, which is clearly insufficient. System design that compensates for human fallibility under pressure.
What This Means For Everyone In Crypto
If you’re a user: your biggest risk isn’t a smart contract exploit. It’s a well-timed, well-crafted message that catches you in a moment of stress, urgency, or distraction. Design your security protocols assuming that moment will happen. Remove single points of human failure.
If you’re building: user education is necessary but not sufficient. Build friction into high-stakes actions. Design for the distracted, pressured, temporarily-fooled user, not the ideal vigilant one.
If you’re in security: the threat model has to include AI-powered social engineering as a primary attack vector, not an edge case. Red team exercises need to include sophisticated AI-assisted social engineering simulations.
If you’re an investor: ask every project you invest in: what’s your human security layer? Not just your smart contract audit. What protects against AI-powered attacks on your team members?
The Real Arms Race
Everyone talks about crypto’s AI arms race as a trading problem. AI trading against AI. Faster algorithms, better predictions.
The real arms race is in security. Attackers using AI to exploit human psychology at scale. Defenders using AI to detect anomalous behavior and flag suspicious communications.
One side is attacking a fixed vulnerability: human cognitive limitations under pressure.
The other side is defending a moving target: human behavior across thousands of employees, users, and community members.
The attackers have a structural advantage. They only need to succeed once.
The defenders need to succeed every time.
That asymmetry is the actual security crisis in crypto. Not the code. The people.
