AI Is Helping Discover Tech Vulnerabilities—And Zcash Is Just the Latest Example – Decrypt

The most recent era of frontier AI fashions are now not simply chatting with customers, producing pictures, or writing code. Researchers are more and more utilizing methods akin to Anthropic’s Claude Mythos and Claude Opus 4.8 and OpenAI’s GPT-5.5 to establish software program vulnerabilities, elevating considerations about what occurs when these capabilities grow to be extensively accessible.

Crypto traders bought a wake-up name concerning the rising risk from highly effective AI this week when Zcash builders disclosed that Claude Opus 4.8 helped uncover a essential vulnerability that would’ve enabled an attacker to mint limitless ZEC. Because of the community’s design, there is not any present approach to know for positive whether or not counterfeit ZEC was, in truth, minted—and that uncertainty led to the value of ZEC crashing late this week.

Specialists warn that many extra vulnerabilities might be discovered within the coming weeks and months as AI software program will get extra succesful—and people instruments grow to be extra accessible. This is a take a look at the rising risk, and the way it’s already impacted the crypto world.



Early AI fashions have been professionally used as coding assistants, serving to builders write, clarify, and debug software program. Because the expertise improved, researchers started utilizing the identical methods for code assessment, software program auditing, and vulnerability analysis.

The transition from coding assistant to safety instrument coincided with a broader shift in how AI was getting used inside software program improvement. After the launch of Claude Code in 2025, Anthropic reported a pointy improve in AI-generated code throughout its engineering groups, reflecting a transfer from fashions that advised code to methods able to writing and working it.

Safety professionals say the implications lengthen past serving to builders write code.

“AI is much better at reviewing code than most individuals and discovering potential vulnerabilities in it,” Danny Jenkins, CEO and co-founder of ThreatLocker, instructed Decrypt. Jenkins mentioned present AI methods are already accelerating vulnerability discovery, whereas newer fashions akin to Mythos might considerably broaden these capabilities, calling it an imminent “massive downside.”

“Will probably be solely a matter of time till somebody unhealthy will get entry to it,” he mentioned.

In response to Jenkins, AI can be reducing the obstacles to entry for vulnerability analysis, permitting extra folks to research code, establish weaknesses, and develop exploits. As entry to more and more succesful methods expands, he expects the tempo of vulnerability discovery to extend.

“Pre-AI, cybersecurity threats and exploits have been rising yearly,” he mentioned. “Submit-AI, it is grow to be even quicker, and I believe it is grow to be quicker for 2 causes. One is which you could now use AI to assist discover vulnerabilities and exploits, and the quantity of people that have the power to do that has massively grown. You do not have to be a script kiddie now.”

As AI methods grew to become extra succesful, firms started making use of them to cybersecurity. On Tuesday, Anthropic expanded entry to Challenge Glasswing, giving 150 firms and establishments entry to Claude Mythos to assist establish and remediate software program vulnerabilities earlier than the mannequin is launched extra broadly.

In April, Mozilla later disclosed that Anthropic’s fashions helped establish tons of of vulnerabilities that it fastened within the Firefox net browser, whereas researchers at Calif used Mythos Preview throughout work that produced one of many first public exploits concentrating on Apple’s M5 chips.

Stanislav Fort, a former researcher at Google DeepMind and Anthropic and now founder and chief scientist of safety agency Aisle, mentioned considerations about AI-powered vulnerability discovery are legitimate, however typically misunderstood.

“The naive response is to attempt to gatekeep entry to highly effective fashions. I believe that is basically safety by obscurity, and safety by obscurity is likely one of the worst concepts within the subject,” Fort instructed Decrypt. “The potential for zero-day discovery is already extensively distributed throughout fashions that nobody can prohibit. Making an attempt to bottle it up on the frontier does not remove the danger; it simply delays it whereas additionally slowing down the defenders who want these instruments most.”

Fort mentioned the higher threat is that defenders, significantly open-source maintainers, might lack entry to the identical superior AI instruments accessible to attackers.

“That imbalance is the true hazard,” he mentioned. “The reply is not restriction; it is democratization of the defensive stack.”

Anthropic shouldn’t be alone in pushing AI fashions aimed toward cybersecurity. In Might, Microsoft launched MDASH, an agentic vulnerability discovery system that the corporate mentioned helped establish beforehand unknown Home windows vulnerabilities.

The chance to crypto

Crypto and DeFi are beginning to really feel the impression of AI-powered bug searching. Blockchain tasks have at all times been enticing targets as a result of there’s some huge cash at stake and far of the code is publicly accessible. Jenkins mentioned as AI will get higher at discovering software program flaws, open-source crypto tasks might grow to be simpler targets for each safety researchers in search of bugs and attackers trying to exploit them.

In one of many clearest examples of how superior AI fashions can assist researchers uncover vulnerabilities that had survived years of human assessment, impartial safety researcher Taylor Hornby disclosed the essential vulnerability in Zcash’s Orchard privateness pool that he found with the help of Claude Opus 4.8.

The flaw might have allowed an attacker to create limitless counterfeit ZEC, and had gone undetected for years earlier than being patched. Whether or not the exploit was really used presently stays unknown.

“The vulnerability was current from Orchard’s activation in Might 2022 till the emergency repair was deployed on June 1, 2026,” Shielded Labs, the group behind Zcash improvement, wrote in a disclosure submit. “Because of the privateness properties of Orchard and the character of the bug, there isn’t any definitive approach to decide, utilizing solely cryptography, whether or not such exploitation occurred.”

The assault comes as DeFi protocols are already dealing with one in all their worst years for exploits. Greater than $840 million was stolen from DeFi tasks within the first 5 months of 2026, together with greater than $600 million in April alone throughout assaults on tasks together with KelpDAO, and Drift Protocol.

The rise of so-called ‘vibe hacking,’ the place attackers use AI coding brokers to automate reconnaissance, credential theft, malware improvement, and different duties, has raised considerations that AI is reducing the obstacles to finishing up subtle cyberattacks

In response to Natalie Newson, senior blockchain investigator at Web3 safety platform CertiK, whereas April was unusually extreme for crypto exploits, the broader development stays extra steady and beneath the height variety of incidents seen in previous years.

“April 2026 was a nasty month for crypto exploits; there have been solely three days with out an exploit during which no less than $10,000 was taken,” she mentioned. “Nonetheless, once we check out the broader image, the variety of incidents (excluding phishing) has arguably been pretty constant and nonetheless decrease than a peak in 2023.”

Whereas AI is making DeFi exploits simpler to hold out, based on Blockaid CTO Raz Niv, the larger threat shouldn’t be AI changing hackers however amplifying them, permitting attackers to deal with extra subtle methods whereas AI handles routine duties.

“The excellent news is defenders can use the identical instruments,” he mentioned. “AI-assisted monitoring and simulation is changing into important for safety groups making an attempt to maintain tempo.”