In this article5 sections
Near AI has disclosed a security incident in which an attacker used a remote code execution flaw in the open-source SGLang inference engine to run commands on one of its servers and move into other internal systems, according to Decrypt. The company’s account, published on November 9, says the intruder downloaded names and email addresses for roughly 100 users and pulled chat histories, settings and logs from 18 agents on an older deployment.
The same week, crypto.news reported that Binance was pushing back on a separate claim by the threat actor ShinyHunters, which advertised 335,900,000 lines of purported Binance user records. No link between the two incidents has been reported.
Also readGnosis Pay Reports Data Breach, Adds 3D Secure Until Dec. 20
Key facts
- Near AI attributed the intrusion to a remote code execution vulnerability in SGLang and traced it to an internal system tied to Private Chat, a feature the company had already retired.
- Approximately 100 users had names and email addresses taken, with subscription plan and account balance information exposed for some of them.
- Data from 18 agents was retrieved from an older Near AI Agents deployment, including chat histories, settings and logs; those agents did not run inside the confidential computing environment.
- Logs also contained NEAR AI Cloud API keys belonging to two customers, and Near AI said both keys have been revoked.
- Binance, in the crypto.news report, said its preliminary review of the advertised dataset found no evidence of a new compromise of its systems or customer data, and a spokesperson cited plaintext passwords, external wallet private keys and jurisdictions where the exchange no longer operates as inconsistencies.
What the confidential-mode question means for users
Near AI sells privacy as a product feature, using trusted execution environments — hardware enclaves that keep prompts encrypted and separated from the operators of the servers, even while a model processes them. The company acknowledged that such isolation “does not protect against a vulnerability in the software running inside the environment,” and said the SGLang flaw let the attacker execute code in the place where confidential requests were handled.
The firm said it has no indication that stored Private Chat conversations or uploaded files were accessed. Its investigation did, however, flag that requests processed in confidential mode on November 8 may have been visible to the attacker. The confirmed losses were account records and legacy agent data; live confidential traffic remains a possibility the review has not ruled out.
Also readRevolut Hit With 6,000 XMR Ransom Demand as Hackers Threaten to Sell Customer Data
Near AI reported the flaw to SGLang’s maintainers. The engine had drawn attention before: in July, Carnegie Mellon University’s CERT Coordination Center documented six vulnerabilities in the project, some of which could allow command execution or credential exposure depending on server configuration. Near AI did not name the specific flaw involved, so there is no basis for tying the incident to those earlier disclosures.
In October, Near AI announced that Brave’s Leo browser assistant was using its confidential inference infrastructure for a supported model. The incident report does not list Brave users among those affected.
Near AI’s response and Binance’s separate claim
In its security incident report, Near AI said it revoked all active sessions and administrative credentials, replaced agent tokens, took Private Chat offline and patched the SGLang vulnerability across every inference server. The company said it has finished rotating the remaining credentials and a full forensic review, has notified users whose records or agents were confirmed accessed, and will make any regulatory notifications that are required. It also contacted the owners of all 18 agents individually and warned users to treat unsolicited messages about their Near AI accounts or balances with suspicion.
The Binance matter is different in kind: it concerns an alleged theft of customer records and credentials rather than a confirmed intrusion. According to crypto.news, ShinyHunters demanded $900,000 in Bitcoin or Monero and said the exchange had “two options left: pay or leak.” ThreatMon, which surfaced the forum listing, noted that the claim had not been independently verified. Binance’s spokesperson suggested the advertised data more closely resembled information harvested by infostealer malware from individual devices than records taken from the exchange.
The exchange’s statement follows a confirmed hot wallet attack on October 29, in which about $6.34 billion was taken; Binance later reported that roughly $2.85 billion had been frozen or intercepted and used company resources to cover affected customer assets. It requested the return of 8,425.997 ETH and about 74.56 million ARB on November 5.
Why it matters
Both cases land on the same weak point: the trust users place in a provider’s claim that sensitive material is walled off. For Near AI, the lesson is that hardware isolation says nothing about the code running inside the enclave, which is exactly the layer the SGLang bug exploited. Whether an attacker reached live confidential prompts or only account metadata changes what customers need to do next. The Binance episode shows the opposite risk — a publicised data-theft claim that may turn out to be recycled malware logs — and how quickly such listings spread before verification. For readers holding accounts at either service, the practical exposure so far is identity and contact information rather than funds, though Near AI has not closed the door on confidential request data.
What to watch
Two things will settle these stories. Near AI’s forensic review may determine whether confidential-mode requests from November 8 were in fact read, and any required regulatory filings will be the public record of that finding. For Binance, the open question is whether ShinyHunters produces a verifiable sample linking the dataset to the exchange’s own systems; ThreatMon has already cautioned that the hot-wallet and private-key claims were unverified.
{{DISCLAIMER}}
Sources: Decrypt, crypto.news
