Immunefi CEO warns frontier AI is accelerating DeFi exploit risk in 2026
April losses topped $634M, and a $290M–$293M Kelp DAO bridge drain exposed single-verifier design risk.
Immunefi CEO Mitchell Amador says frontier AI models are shifting the offense-defense balance in crypto security toward attackers, driving a 2026 resurgence in DeFi exploits. He argues the industry is entering a multi-year “survival period” where exploit headlines can remain a recurring market risk factor.
Key Takeaways
- More than $634 million was stolen from crypto platforms in April 2026, the highest monthly total since February 2025, based on DefiLlama’s hack-loss aggregation.
- Immunefi CEO Mitchell Amador described the current environment as a “vulnerability apocalypse,” arguing frontier AI has improved attackers’ ability to find and exploit weaknesses.
- Claude Opus 4.8 and ChatGPT 5.5 were cited as examples of models Amador linked to the 2026 resurgence in hacks.
- The April 19 Kelp DAO rsETH bridge exploit drained about 116,500 rsETH worth roughly $290 million to $293 million, and LayerZero tied the failure mode to a 1/1 DVN single-verifier path.
Immunefi CEO: Frontier AI Has Tilted the Field Toward Attackers
Mitchell Amador, CEO of bug bounty platform Immunefi, is putting a specific narrative on 2026’s exploit tape. He said new frontier AI models have shifted the cybersecurity playing field in favor of attackers, calling it a “vulnerability apocalypse” tied to a resurgence in DeFi hacks.
Amador explicitly pointed to Claude Opus 4.8 and ChatGPT 5.5 as examples of models whose proliferation he linked to the renewed pace of incidents. The packet does not include independent measurement of how much these models contributed to April’s losses versus other drivers, so the AI-causality framing remains an executive assessment rather than a quantified attribution.
His comments came in the wake of Anthropic’s release of its Claude Mythos model, Fable 5, which sparked concerns about whether more capable systems could accelerate exploit development. Anthropic said on Tuesday that Fable 5 includes safeguards that reroute cybersecurity topics to a different model, Claude Opus 4.8. The operational effectiveness of that safeguard is not established in the provided material.
April’s $634M Hack Month and the Prior Peak in February 2025
The market-relevant part is the size of the losses, not the rhetoric. Illicit actors stole more than $634 million from cryptocurrency platforms in April 2026, the highest monthly total since February 2025, according to DefiLlama data.
The prior peak month referenced was February 2025 at roughly $1.4 billion, with the Bybit hack cited as a driver. Details on that Bybit incident are not provided in the packet, but the comparison matters for traders because it frames April as a re-emerging tail-risk regime rather than a one-off anomaly.
In practice, a $634 million month is enough to change positioning behavior. It can widen risk premia on DeFi-adjacent tokens, pressure bridge and restaking flows, and pull liquidity toward venues and protocols perceived as simpler or more battle-tested.
Kelp DAO’s $290M–$293M rsETH Bridge Drain Shows the Single-Point-of-Failure Problem
April’s headline number also contains a familiar lesson. On April 19, an attacker drained about 116,500 restaked Ether (rsETH), worth roughly $290 million to $293 million at the time, from Kelp DAO’s LayerZero-powered rsETH bridge.
LayerZero attributed the incident’s risk profile to Kelp DAO’s 1/1 decentralized verifier network (DVN) setup. In that configuration, cross-chain messages rely on a single verifier path, creating a single point of failure. LayerZero also said it had previously advised against using that setup.
This is the uncomfortable counterpoint to the frontier-AI narrative. Even if attacker tooling is improving, large losses can still come from configuration and verification design choices. For traders, that keeps bridges and cross-chain messaging architecture on the short list of due-diligence checks.
Signals to Watch for Frontier AI blamed for DeFi hack
DefiLlama’s monthly hack-loss totals after April 2026 will be the cleanest scoreboard. The reference points are February 2025 (~$1.4 billion) and April 2026 (> $634 million). If the series stays elevated, exploit risk becomes a persistent input into DeFi pricing rather than an episodic shock.
Amador’s proposed stopgap is broader adoption of “crowdsourced security solutions,” including expanded bug bounties. Follow-through from major protocols on bounty scope and responsiveness will matter more than slogans because it is one of the few levers cited that could compress the timeline to “less than two years.”
On the Kelp DAO front, additional post-mortems and remediation steps will be key, especially any moves away from 1/1 DVN configurations that LayerZero described as a single point of failure.
Anthropic’s next statements about Fable 5’s safeguards also matter at the margin, but the packet provides only a “Tuesday” reference with no specific date or operational detail on how rerouting cybersecurity topics to Claude Opus 4.8 performs under real-world adversarial pressure.
How Traders Should Frame the Next 2–4 Years of DeFi Exploit Risk
I treat Amador’s AI framing as a sentiment catalyst until the incident data starts tying specific exploit patterns to model-enabled workflows. The April figure still does the real work here. A >$634 million month tells traders that exploit risk is back as a live variable for liquidity, bridge usage, and protocol selection, even if the root-cause attribution remains fuzzy.
The threshold that matters is whether the market starts pricing security like a regime, not a headline. If losses remain elevated and protocols respond with measurable changes like bigger bounties and fewer single-verifier bridge designs, the setup starts to look structural rather than narrative-driven, and that is when exploit risk becomes a durable input to DeFi risk premia.
