TheBlock
NEAR AI contained a security breach after an attacker exploited a remote code execution vulnerability in the open-source SGLang inference engine running on its servers. The intruder executed code on one inference server, reached an internal system behind the deprecated Private Chat feature and downloaded account details for approximately 100 users. The investigation also identified access to 18 legacy NEAR AI Agents and logs containing API keys for 2 Cloud customers.
Why it matters
The exposed information included email addresses, names and, for some users, plan and balance data. NEAR AI said the attacker could have observed requests being processed on the compromised server on Nov. 8, including requests made in confidential mode. The company found no indication that stored Private Chat conversations or uploaded files were accessed.
The incident illustrates the boundary of confidential computing. NEAR AI said the technology protects data from the operator and host, but it does not protect against vulnerabilities in software running inside the environment. The deprecated Private Chat system and a legacy agent deployment outside the confidential computing environment expanded the attack surface once code execution was obtained.
Market impact
NEAR AI closed the unauthorized access, replaced affected credentials and patched the vulnerability across all inference servers. It paused NEAR AI Cloud during the investigation before restoring access, revoked active sessions and administrative credentials, rotated agent tokens and took Private Chat offline. The company also completed a forensic review, continued credential rotations and contacted affected users and the owners of the 18 agents individually.
The 2 exposed Cloud API keys were revoked, limiting the risk of continued unauthorized access. The broader signal for AI infrastructure is that confidential-computing guarantees do not remove the need to audit the code, dependencies and deployment configurations operating inside protected environments.
Frequently asked questions
-
What information was exposed in the NEAR AI breach?
The attacker downloaded account details for approximately 100 users, including email addresses, names and, in some cases, plan and balance information. The 2 Cloud customer API keys found in logs were also exposed.
-
Were stored Private Chat conversations or uploaded files accessed?
NEAR AI found no indication that stored Private Chat conversations or uploaded files were accessed. It said requests being processed on the compromised server, including confidential-mode requests, could have been observed on Nov. 8.
-
Were the exposed NEAR AI Cloud API keys still active?
No. NEAR AI revoked the API keys found in logs, replaced affected credentials and rotated active sessions, administrative credentials and agent tokens.
-
What actions did NEAR AI take to contain the breach?
NEAR AI patched SGLang across all inference servers, closed unauthorized access, paused and then restored Cloud, took Private Chat offline and completed a forensic review. It also began contacting affected users and the owners of the 18 agents.
-
What did confidential computing fail to protect against?
It did not protect against a vulnerability in the software running inside the protected environment. NEAR AI said the SGLang flaw allowed the attacker to execute code on the inference server and access requests being processed there.
CryptoSlate
CoinDesk
Crypto News
WuBlockchain
Lookonchain