This article was compiled and organized by BlockWeeks
This week's research brief focuses on three things: a review of the Bitget $387.5 million theft incident, the SEC's frequently asked questions (FAQ) regarding token buybacks, and the key points from the White House AI summit.
1. Lazarus Strikes Again: A Review of the Bitget $387.5 Million Theft
Last week, Bitget disclosed that attackers stole $387.5 million from its hot wallets and warm wallets through a series of unauthorized transfers, and it temporarily suspended all withdrawals. The losses spanned 11 blockchains, including Ethereum, XRP, TRON, Arbitrum, Base, and BSC, with the largest single-chain loss being on XRP, at approximately $83 million.
According to Bitget CEO Gracy Chen, the private keys and cold wallets were not compromised—a claim also corroborated by the hundreds of millions of dollars in tokens still remaining in the attacked addresses. What the attackers actually exploited were two third-party security products (using a zero-day vulnerability to obtain high-level internal credentials), implanting forged withdrawal instructions into the wallet backend and deleting these instructions after the funds were transferred out. SlowMist, the security firm hired by Bitget, traced the earliest malicious activity back to August 31 and reconstructed a custom withdrawal tool built around Bitget's withdrawal logic. Overall, this points to a highly researched, extremely targeted wallet management attack.
Based on IP behavior and on-chain analysis, Bitget attributed the attack to North Korean actors; on-chain tracking firms Elliptic and TRM found overlaps between the relevant wallets and past hacking incidents. Most of the stolen funds were transferred via THORChain, and THORChain refused Chen's request to stop services on the grounds of being "decentralized and permissionless." NEAR Intents took the opposite approach, intercepting more than $50 million in money laundering flows and freezing approximately $503,000 in the process. Bitget's protection fund and the applicable proof of reserves (PoR) overcollateralization covered this loss, with all losses borne by the protection fund, which has been replenished to $300 million.
Then on Thursday, NEAR Intents itself was also attacked: a vulnerability in its Omni deposit/withdrawal infrastructure caused approximately $3.8 million in losses, and officials said they would fully compensate. ZachXBT flagged multiple abnormal outflows from BSC hot wallets, with funds transferred to KuCoin and further bridged to Bitcoin.
2. Private Keys Have Been "Solved"; the Real Target Is the Approval Pipeline
If this incident sounds familiar, it is because in February 2025 the North Korean hacking group Lazarus stole $1.5 billion from Bybit in the same way—they did not steal Bybit's private keys, but instead compromised Safe{Wallet}'s infrastructure, showing signers a tampered interface, after which the signers approved what looked like a routine transfer. The cryptography itself was completely normal; it simply faithfully signed what it was asked to sign.
This is precisely the current landscape of major attacks: no one attacks private key mathematics, because this mathematics (currently) cannot be broken; they attack the people or systems that "tell this mathematics what to do." The private key security of major crypto exchanges is among the best in the world, and the last time a major exchange had its private keys compromised was years ago. The private key attack suffered by Phemex in January 2025 resulted in losses of only about $50 million, and Coincheck's case in 2018 involving more than $500 million is already "ancient history." Private key security has basically been solved, so for attackers, the most valuable target is not the key itself, but the link that determines "what is presented to the key." Bitget has so far not disclosed the names of the vendors exploited, and the forensic report refers to them only as "Product A" and "Product B."
The form of the theft itself also illustrates the problem: if the attackers had obtained the private keys, they would have swept all the funds away; but Bitget's first notification described only unauthorized transfers from a limited number of hot wallets. A slow drain is the fingerprint of "an intermediate service being compromised" rather than "private keys being compromised."
Nick Szabo wrote as early as 2001 in "Trusted Third Parties Are Security Holes." This industry pays a price of roughly nine figures every year to relearn this lesson: the mathematics of cryptography is secure; what needs hardening is the structure around it. This summer, the AI industry learned a similar lesson.
3. SEC Buyback FAQ: The Boundary Between Functional Systems and "No Central Party"
The FAQ released by the SEC provides a more detailed explanation regarding token buybacks. On the "investment contract" issue, promoting the current utility of a crypto system, or promoting its future vision without attaching profit implications, is likely not to constitute a promise of "essential managerial efforts" under the Howey test; once the system is already operational, services provided to secure, maintain, improve, or enhance it, or to promote network effects, also do not constitute essential managerial efforts.
The most closely watched part is the buyback section. FAQ Q2.5 asks: whether an issuer's announcement of a buyback plan for non-security crypto assets (whether for treasury management, supply reduction, protocol-funded burns, or rebalancing) itself constitutes a promise of "essential managerial efforts"—which is one of the paths by which a token sale could fall under a Howey investment contract. The staff's answer is: when the crypto system is already operational and there is no central party, such an announcement does not constitute that kind of promise.
It is worth noting that the phrase "and there is no central party" was only added on September 28, 2026—three days after the FAQ was released, after industry criticism that the original wording (requiring only that it be "already operational") was too broad. If the system is not yet operational, then as long as the issuer packages the buyback as bringing benefits or returns to token holders, the announcement may still constitute such a promise. However, the FAQ does not explain how buyback announcements should be handled in systems that are "already operational but have a central party"—since the answer no longer covers that scenario, it will likely fall back to the general facts-and-circumstances Howey analysis.
