AI-driven hacks
Crypto hacks and drains where AI did the attacking, made the lure, wrote the malware, or was the way in. An incident is listed here only when a source says AI was involved.
16 AI-driven hacks in the last 12 months
16 AI-driven hacks in the last 12 months
15 AI-driven hacks in 2026
11 AI-driven hacks in 2025
- AI-driven incidents
- 26
- Live now
- 5
- Critical
- 5
- Days since the last one
- 13
- Longest quiet run: 85 days
- AI did the attacking1AI models or agents, sometimes swarms of them, picked the targets, found the bugs, wrote the exploits or ran the attack.
- AI-made lures7Deepfake video or voice, AI-written phishing, AI-generated fake personas, sites and support bots.
- AI-built malware6Stealers, drainers and malicious packages written or adapted with AI, malware that turns AI tools against you, and fake packages named after AI mistakes.
- AI agents as the target12An AI agent, trading bot, agent plugin or AI coding tool was the way in: prompt or memory injection, malicious skills and MCP servers, agents leaking keys.
AI did the attacking 1
Stay safe: Projects: assume contracts and front-ends are scanned around the clock, so audit, cap admin powers and monitor. Users: keep funds out of new or unaudited contracts.
-
Drift Protocol drained of $270M+ after signers pre-signed durable-nonce admin transactions
On Apr 1, 2026, an attacker took admin control of the Solana perps exchange Drift and drained its vaults of at least $270 million; SlowMist lists $285 million. Weeks earlier, Security Council signers had been led into 'unauthorized or misrepresented' approvals. Because those transactions used durable nonces, they stayed valid until the attacker chose to submit them. The funds were bridged to Ethereum, including more than $230 million in USDC via Circle's CCTP. No code bug was needed. The lesson for teams is that any durable-nonce signing request is high risk.
Drift Protocol (Security Council multisig)·Solana·$285M stolen·Since Apr 1, 2026·
AI-made lures 7
Stay safe: Check anyone who asks for money, a signature or a download on a second channel you already trust. A familiar face or voice on a call is no longer proof.
-
Joint advisory: North Korea's WaterPlum fake-job campaign infected 30,000 devices, hit 7,000+ wallets
On 18 September 2026 Japanese, US, Australian and German authorities warned that the North Korean group WaterPlum, linked to the 'Contagious Interview' campaign, infected at least 30,000 devices in more than 100 countries between December 2025 and July 2026. Posing as AI, crypto or NFT firms and recruiters, they get targets to run code during fake interviews and coding tests, including malicious npm packages and VS Code projects, then steal private keys, seed phrases and passwords. Funds or credentials were taken from over 7,000 wallets, and more than 1.7 billion yen (about $10.71M) went to North Korea. The advisory does not say the campaign has stopped.
Crypto, AI and NFT developers and job seekers·Multiple EVM chains·$11M stolen·Since Sep 18, 2026·
-
North Korea's UNC1069 used an apparent deepfake CEO on a fake Zoom call to infect a crypto fintech
Mandiant reported on 9 February 2026 an intrusion at a crypto-sector fintech that began with a message from a crypto executive's hijacked Telegram account and a Calendly invite to a spoofed Zoom meeting. On the call the victim saw what appeared to be a deepfake of another crypto company's CEO, was told their audio was broken and was talked into running 'troubleshooting' commands (ClickFix). Those commands installed seven malware families on macOS that collect credentials, browser and Telegram data to enable crypto theft. Mandiant tied it to the North Korea-linked group UNC1069 and reported no stolen amount.
Crypto and DeFi fintech staff on macOS·Multiple EVM chains·Since Feb 9, 2026·
-
THORChain co-founder loses about $1.35M after a fake Zoom call showing a deepfake of a friend
In early September 2025 THORChain co-founder John-Paul Thorbjornsen got a Zoom link from a friend's hacked Telegram account, joined in his browser and, by his account, saw a deepfake of his friend with no sound before leaving. An old MetaMask wallet of his was later drained of about $1.35M in Kyber Network and THORSwap tokens (PeckShield valued it at about $1.2M); he believes the key was reached through iCloud Keychain or a zero-day. ZachXBT attributed the theft to North Korean hackers, and THORChain said the protocol itself was not affected.
THORChain co-founder's personal wallet·Ethereum·$1.4M stolen·Since Sep 4, 2025·
-
YouTube 'MEV bot' tutorials, some AI-generated, tricked users into deploying drainer contracts; $900K+ taken
SentinelLABS reported in August 2025 that scammers use YouTube tutorials to get people to deploy a 'free' Ethereum arbitrage bot contract through Remix, fund it with ETH and press Start. Hidden code sends the ETH to an attacker wallet, and the top attacker address alone received 244.9 ETH (about $902,000). The videos ran on aged channels with manipulated comments, and SentinelLABS says several appear to be AI-generated. Versions of the scam have run since at least early 2024 and were described as ongoing.
Ethereum users following YouTube 'MEV bot' tutorials·Ethereum·$900K stolen·Since Aug 5, 2025·
-
Phishing campaign used the AI site builder Lovable to fake Aave and push wallet connections
Proofpoint reported on 20 August 2025 that criminals were using Lovable, an AI website builder that makes sites from text prompts, to create and host phishing and wallet-drainer pages. In June 2025 nearly 10,000 emails sent through SendGrid led to a Lovable-built app that redirected to a fake Aave site asking people to connect a crypto wallet; Proofpoint said the likely aim was to steal assets from any connected wallet. Lovable removed a phishing cluster and added real-time checks after the report. No loss figures were published.
Aave and other DeFi users receiving fake reward emails·Multiple EVM chains·Since Jun 17, 2025·
-
North Korea's BlueNoroff used deepfaked executives on a Zoom call to infect a crypto foundation employee's Mac
Huntress reported on 18 June 2025 that an employee of a cryptocurrency foundation was contacted on Telegram, sent a Calendly invite and led to a fake Zoom domain. Weeks later the employee joined a group call that contained several deepfakes of the company's senior leadership; when 'audio problems' came up, they were told to install a Zoom extension that was really a malicious AppleScript. It installed a set of macOS backdoors, a keylogger and screen recorder, and a crypto-wallet stealer called CryptoBot. Huntress did not say whether funds were stolen.
Crypto foundation staff on macOS·Multiple EVM chains·Since Jun 11, 2025·
-
Deepfake Zoom calls of Polygon's co-founder push malware to startup founders from a hijacked Telegram account
On 13 May 2025 Polygon co-founder Sandeep Nailwal warned that attackers had taken over the Telegram account of the head of Polygon Ventures and were messaging startup founders in its network. Targets were invited to Zoom calls showing AI-generated video of Nailwal and other team members with no audio, and were urged to install an 'SDK' to fix the sound, which would infect their computers. No losses were reported. Do not install software during a call, even when the people on screen look like contacts you know.
Startup founders in Polygon's investment network·Multiple EVM chains·Since May 13, 2025·
AI-built malware 6
Stay safe: Check that a package, extension or app is the real one before installing it, even when an AI assistant suggested it.
-
Noir wallet drainer kit hides its backend behind DNS TXT records and DoH
Infoblox Threat Intel described Noir on 9 October 2026, a wallet-drainer kit that looks up its backend pool through DNS TXT records over public DNS-over-HTTPS, so operators can repoint every lure page by changing one record. Lures impersonate CoinMarketCap, DexScreener, OKX, Uniswap, Lido, Hyperliquid, Morpho, Ondo, LayerZero, MegaETH, Polymarket and others with fake votes, airdrop checks and swap clones. It steals by getting token spending permissions, then cuts the WalletConnect session to hide the trail. Infoblox gave no loss figures and says the kit was likely AI-developed.
Users of Uniswap, Lido, Hyperliquid, Polymarket and other brands targeted by Noir lure pages·Multiple EVM chains·Since Oct 9, 2026·
-
Arkanix stealer, likely built with LLM help, targeted crypto wallets before its makers shut it down
Kaspersky analysed Arkanix, a malware-as-a-service info-stealer advertised on dark-web forums and Discord from October 2025, as Security Affairs reported on 24 February 2026. It took crypto wallet data, with paid builds offering injection into Exodus and Atomic wallets, along with browser passwords, cookies, Telegram sessions and Discord logins. Kaspersky says the code shows probable traces of LLM-assisted development, and the operators shut the project down about two months after launch. No victim counts or losses were published.
People infected with the Arkanix stealer (Exodus, Atomic and browser wallet users)·Multiple EVM chains·Since Feb 24, 2026·
-
North Korea-linked KONNI targets blockchain developers with an AI-generated PowerShell backdoor
Check Point Research reported on 22 January 2026 that the North Korea-linked group KONNI, long focused on South Korean government targets, is now going after developers and engineering teams on blockchain and crypto projects. Phishing lures dressed up as project documentation deliver a PowerShell backdoor that Check Point calls AI-generated, with activity seen in Japan, Australia and India. A compromised developer can expose source code, infrastructure and blockchain credentials. No losses were reported.
Blockchain and crypto developers and engineering teams (Asia-Pacific)·Multiple EVM chains·Since Jan 22, 2026·
-
Kaspersky: North Korea's BlueNoroff uses generative AI to speed up malware in GhostCall and GhostHire crypto attacks
Kaspersky disclosed on 28 October 2025 two BlueNoroff campaigns aimed at Web3 and crypto organisations in India, Turkey, Australia and other countries in Europe and Asia, active since at least April 2025. GhostCall reaches executives and founders on Telegram with fake investor meetings, where an 'audio fix' installs malware; GhostHire sends blockchain developers fake job tests in GitHub repositories that hide malware. Kaspersky says generative AI has helped the group develop malware faster, and the payloads include crypto stealers. No loss figures were given.
Web3 and crypto executives, founders and blockchain developers·Multiple EVM chains·Since Oct 28, 2025·
-
Compromised Nx npm packages ran victims' AI command-line tools to hunt for wallet and keystore files
Around 26 August 2025 malicious versions of Nx, a widely used JavaScript build tool, were published to npm and stole GitHub and npm tokens on install. The malware also looked for installed AI command-line tools (Claude, Gemini and Amazon Q) and ran them to search the disk for crypto wallet and keystore files, .env files and SSH keys, then published the results to repos in the victim's own GitHub account. Wiz counted over 1,700 users with leaked secrets and over 6,700 private repos made public, but saw no successful crypto theft, partly because the AI tools often refused. Anyone who installed the bad versions should rotate tokens and move funds from wallet files on that machine.
Nx build system on npm·Multiple EVM chains·Since Aug 26, 2025·
-
AI-written npm package @kodane/patch-manager drained Solana wallets after 1,500+ downloads
On 28 July 2025 an npm package called @kodane/patch-manager, posing as a license-validation and registry cache tool for Node.js, was published with a hidden wallet drainer in its install script. Once installed it looked for crypto wallets on Windows, macOS and Linux and sent the funds to a Solana address, leaving just enough for fees. It was downloaded more than 1,500 times before npm flagged it on 30 July; Safety, which found it, said the code appears to be largely AI-generated. No total loss was published.
npm package @kodane/patch-manager·Solana·Since Jul 28, 2025·
AI agents as the target 12
Stay safe: Give agents a small, capped wallet with allow-listed destinations, and never let an agent that can sign read untrusted content.
-
Knostic: 'gadgethumans-mcp' npm MCP server sends users' raw wallet private key to a remote server
Knostic reported on 8 September 2026 that gadgethumans-mcp, an npm MCP server that promises AI agents automatic x402 micropayments, asks users to set a wallet private key but does no local signing. Instead it copies the raw key into a request header and sends it to a remote endpoint on every recognized tool call, which would let whoever receives it take over the wallet. The package had 2,327 downloads in July and August 2026 and was still on npm when the report came out; Knostic found no confirmed thefts. Anyone who set a key in it should move funds and use a fresh, low-balance wallet.
gadgethumans-mcp (npm)·Base·Since Sep 8, 2026·
-
Malicious OpenClaw AI-agent skills pose as Solana and Phantom tools to steal wallet keys
Bitdefender found that about 17% of the OpenClaw (ClawHub) AI-agent skills it analyzed in the first week of February 2026 behaved maliciously. Solana-themed skills made up 9.3% of the malicious set and Phantom-wallet skills 8.2%. Some skills told users to run fake installers that dropped the AMOS macOS stealer. Solana skills that read SOLANA_KEYPAIR_PATH can expose the plain-text keypair file that holds a wallet's private key. Payloads were staged on 91.92.242[.]30 and an impersonated OpenClaw GitHub repo. Anyone who installed crypto skills and ran their setup commands should move funds to a new wallet.
OpenClaw (ClawHub) skill users·Solana, Multiple EVM chains·Since Feb 5, 2026·
-
Zscaler: live web campaigns hide prompt injections telling browsing AI agents to make crypto payments
SecurityWeek reported on 6 July 2026 that Zscaler found two campaigns in use that plant hidden instructions on web pages for AI agents that can browse and pay. One used SEO poisoning to lure agents searching for a fake Python package, requests-secure-v2, backed by 10 GitHub repositories; the other typosquatted the DeFi portfolio tracker DeBank. In Zscaler's own tests 4 of 26 models were manipulated into making a payment and 2 mistook the fake DeBank site for the real one. No real losses were reported, but agents with spending power should only pay allow-listed recipients.
AI agents that browse the web and can pay·Multiple EVM chains·Since Jul 6, 2026·
-
TrapDoor packages on npm, PyPI and Crates.io steal wallet data and plant hidden prompts for AI coding tools
Socket reported in late May 2026 a campaign it named TrapDoor: more than 34 malicious packages and 384+ related versions across npm, PyPI and Crates.io that steal crypto wallet data and other developer secrets. The packages also tried to alter files that AI coding tools read, such as .cursorrules and CLAUDE.md, using hidden Unicode instructions meant to get the assistant to run scan-like steps that find and send out secrets. No confirmed losses or victim counts have been published. Developers who installed flagged packages should rotate credentials and move wallet funds.
Developers installing packages from npm, PyPI and Crates.io·Multiple EVM chains·Since May 26, 2026·
-
Encoded X post tricks Grok and Bankrbot into sending about 3 billion DRB tokens from Grok's wallet
On 4 May 2026 an attacker drained about 3 billion DRB tokens, worth roughly $150,000 to $200,000 depending on the estimate, from a Bankr wallet linked to xAI's Grok on Base. Reports say the attacker first sent the wallet a free Bankr Club membership NFT that widened what it could authorize, then posted a Morse-coded message; Grok decoded it and the Bankr automation treated Grok's reply as a valid transfer order. No private key, smart contract or malware was involved, and some funds were later returned. An AI's text output should never count as permission to move money.
Grok-linked Bankr wallet (Bankrbot on X)·Base·$155K stolen·Since May 4, 2026·
-
Solana wallet MCP server on PyPI sent users' private keys to a remote server
An advisory published on 31 March 2026 flagged latinum-wallet-mcp, a PyPI package that runs a Solana wallet as an MCP server for AI agents, as malicious. From version 0.0.32 it sent the wallet's private key, along with other metrics, to a remote server during wallet setup, and from 0.0.34 it added itself as an MCP server for Claude on import; none of this was in the public GitHub code. Versions 0.0.32 to 0.0.36 are listed as affected. Anyone who installed them should treat that Solana key as stolen and move funds to a new wallet.
latinum-wallet-mcp (PyPI)·Solana·Since Mar 31, 2026·
-
Compromised LiteLLM releases on PyPI steal cloud keys and crypto wallet data from AI developers
On 24 March 2026 attackers tracked as TeamPCP published LiteLLM 1.82.7 and 1.82.8 to PyPI with hidden code that runs on import, and in 1.82.8 whenever Python starts. LiteLLM is a popular open-source gateway that AI apps and agents use to reach many LLM providers, with over 3.4 million downloads a day. The payload collected SSH keys, cloud credentials, Kubernetes secrets, .env files and crypto wallet data and sent them to a lookalike domain. Both versions were removed; anyone who installed them should rotate secrets and move funds from wallets on affected machines.
LiteLLM on PyPI·Multiple EVM chains·Since Mar 24, 2026·
-
Claude-assisted commit adds a North Korea-linked wallet-stealing npm package to an AI crypto trading agent
ReversingLabs reported on 29 April 2026 that a 28 February commit to openpaw-graveyard, an autonomous crypto trading agent built for Solana, was co-authored by Claude Opus and added a bait package (@solana-launchpad/sdk) that silently pulled in the malicious @validate-sdk/v2. The payload steals .env and JSON files and crypto wallet credentials, and adds attacker SSH keys for remote access. ReversingLabs ties the campaign, 60+ packages and 300+ versions over about seven months, to North Korea's Famous Chollima and believes it deliberately targets AI coding agents. No losses were reported; check lockfiles for the named packages and rotate keys on affected machines.
AI trading agent projects and Web3 developers on npm·Solana·Since Feb 28, 2026·
-
AI trading agent Lobstar Wilde sends about $250K in tokens to an X user who asked it for 4 SOL
On 22 February 2026 Lobstar Wilde, an experimental AI trading agent with its own Solana wallet built by an OpenAI developer, answered an X reply asking for 4 SOL for a relative's medical treatment. Instead of a small amount it sent about 53 million LOBSTAR tokens, roughly 5% of supply and about $250,000 at the time, and the recipient sold them. The bot called it an accident; a decimal error was suggested by an X user but not confirmed by the creator. Agents that can send funds in response to public replies need hard per-transfer caps and human approval.
Lobstar Wilde, an experimental AI trading agent on X·Solana·$250K stolen·Since Feb 22, 2026·
-
Malicious 'contractshark.solidity-lang' Cursor extension steals an Ethereum core developer's private key
Cointelegraph reported on 13 August 2025 that an Ethereum core developer's hot wallet was drained after he installed 'contractshark.solidity-lang', a fake Solidity extension for the Cursor AI code editor with over 54,000 downloads. The extension read his .env file and sent the private key to the attacker, who waited three days before emptying the wallet. The loss was a few hundred dollars because he kept only test funds there and his main holdings on hardware wallets. Keep keys out of .env files on development machines and use small, separate test wallets.
Cursor AI code editor users·Ethereum·Since Aug 13, 2025·
-
Fake Solidity extension for the Cursor AI code editor leads to about $500K crypto theft from a developer
Kaspersky reported on 10 July 2025 that a blockchain developer lost about $500,000 in crypto after installing a fake Solidity extension in Cursor, an AI code editor built on VS Code that gets extensions from the Open VSX registry. The extension gave no Solidity support; it installed ScreenConnect remote-access software, then the Quasar backdoor and a stealer that took browser, email and wallet data, including seed phrases. The attackers used the seed phrases to drain the developer's wallets. Developers should vet editor extensions and never keep seed phrases or funded keys on a coding machine.
Cursor AI code editor users (Open VSX extensions)·Multiple EVM chains·$500K stolen·Since Jul 10, 2025·
-
Attacker breaches AIXBT AI agent's dashboard and queues prompts that send 55.5 ETH from its wallet
On 18 March 2025 an attacker got into the dashboard of AIXBT, an autonomous AI crypto agent, and queued two prompts that made it transfer about 55.5 ETH (about $106,200) out of its wallet on Base. The maintainer said core systems were not affected and the AI itself was not manipulated; the weak point was access to the agent's control dashboard. Keys were swapped, dashboard access was paused and the attacker's addresses were reported to exchanges. Anyone running an AI agent with a wallet should lock down its control panel and cap what it can send.
AIXBT AI agent wallet·Base·$106K stolen·Since Mar 18, 2025·
No drains match these filters.