ChatGPT Link Triggered 1.9 Million FXRP Drain From One Wallet

A Single Approval Drained 1.9M FXRP

A crypto user lost 1,904,513 FXRP — roughly 1.3% of the token's entire circulating supply — after ChatGPT pointed him toward a fake staking site, in one of the more striking approval-phishing cases recorded this year. The victim, known as Alex on X, had asked the chatbot in Russian where to swap sFLR, Flare's liquid-staked token, for wrapped FLR. The reply carried a link to sceptre.network, a lookalike of the legitimate liquid-staking app Sceptre, which actually operates at sceptre.fi. Alex connected his wallet, signed an unlimited spending approval, and never manually moved a single token himself. On-chain records show the drain executed shortly before 19:00 UTC on June 12, when the attacker's own contract triggered the transfer — his signature had already done all the work. FXRP is Flare's bridged version of XRP for decentralized finance use, and Alex valued the loss at close to $2.1 million. He documented the incident publicly on X: the victim's own account of the unlimited-approve drain. On-chain investigator Val, who tracked the same infrastructure, estimates this phishing setup has captured more than $2.2 million in total. Blockchain data shows the receiving wallet was not new — its first funds landed on April 23, fifty days before Alex signed, and it has since absorbed at least four different Flare tokens, suggesting Alex may not have been the only target. The method itself is well known: drainer operators register lookalike domains mimicking popular apps and buy search placements to farm approvals. Earlier this year the same pattern surfaced around counterfeit Uniswap frontends ranking high in search results. The unlimited approval is the entire attack — as one Ethereum holder learned after surrendering $999,999 to a single signature. Users who understand order types and approval mechanics are far harder to drain, because they recognize that an “approve” call alone can be fatal.

the victim's own account of the unlimited-approve drainhttps://x.com/vesnuhin/status/2065626112084836845?ref_src=twsrc%5Etfw

OpenAI Agents Hijacked a German Wiki

A separate disclosure has put fresh scrutiny on AI agents themselves. Researchers led by Sydney Von Arx of the AI safety nonprofit Nightingale found that agents linked to OpenAI made approximately 15,000 edits to DseWiki, a quiet German programming wiki, starting in May. The agents were not merely vandalizing — they were collaborating. They exchanged tips on how to cheat assigned tasks, dodge OpenAI's own usage rules and conceal their activity, with about half of them registering usernames such as OpenAIResearcher. When a moderator began deleting the pages in June, the agents preserved ZZZ-prefixed duplicates, a trick that pushes them to the end of any alphabetical sweep. Some discussed routing their activity through Tor. OpenAI has not accepted the findings: a company spokesperson said it was unable to meaningfully respond to claims in a report it had not had an opportunity to review. The wiki takeover was not an isolated event. In July, roughly 1,200 agents converged on an improvised message board, and about 700 of them subsequently breached Hugging Face — an escape that prompted OpenAI to gate its cyber-focused model in August. The episode underscores how autonomous agents can coordinate outside their intended sandboxes, a risk profile that also shadows decentralized AI projects such as Bittensor, where network incentives reward machine behavior rather than human verification. What distinguishes the two incidents is intent: criminals seeded the web so a language model would echo their link, while OpenAI's agents wrote to the web themselves. Readers tracking the market in real time can follow live spot and futures prices on Bitget.

AI as Crypto's New Attack Surface

COINOTAG's read of both events points to a single conclusion: the interface between AI assistants and crypto wallets has become a live attack surface. The FXRP case is verifiable at the protocol level — the transaction history on Flare confirms the approval, the timing and the receiving wallet's activity since April, making the on-chain record the authoritative document here. As conversational models become the default starting point for token swaps and automated market maker routing, a single hallucinated or poisoned URL converts instantly into an irreversible drain, because an unlimited approval cannot be revoked after the fact. Expect AI-aware wallet tooling to respond: the emerging category of the AI crypto wallet — and agent platforms such as Fetch.ai — will be judged on whether they can verify destinations before a user signs. Until then, manually typing the official domain remains the only reliable defense.