Follow us for the latest model releases, platform updates, industry developments, and practical engineering insights.
AI has reduced the cost of finding potential vulnerabilities while shifting the burden of verification onto vendors. Faced with a surge in low-quality reports, Apple has begun limiting the number of submissions and imposing a 30-day cooldown period, but rate limits may also keep genuine critical vulnerabilities from getting through.
The Yangtze River Delta (Jiaxing) Token Operations Center was recently launched, aggregating more than 100 large models through unified APIs, metering, settlement, and security audits. What truly deserves attention is not the number of models, but the fact that local computing resources, fiscal policies, and enterprise AI procurement are beginning to be integrated into a single operational framework.
Truffle Security scanned approximately 7.6 PB of data on Hugging Face, exposing the systemic risk that training corpora have long contained cloud credentials, API keys, and private keys. The problem is not limited to data leakage; these secrets may also be memorized, replicated, and reproduced by models.
Cursor recently removed cost information from the usage page and CSV exports. Although token or request counts still reflect usage volume, they cannot replace actual monetary amounts, directly affecting team budget alerts, project cost allocation, and anomaly investigation.
