Today's AI trends
OpenAI Agents SDK v0.22.2 fixes UnixLocal file-API symlink races. Its sandbox boundary material explicitly treats host-path materialization, symlink swaps, traversal, and archive extraction as filesystem trust-boundary concerns. That matters for agents with persistent workspaces and file operations: process isolation alone is not enough if files cross between host and sandbox. Practical implication: add symlink-swap, traversal, archive, and host-path cases to the security regression suite for an Agent-Runtime. v0.22.2 release Sandbox runtime boundary
vLLM 0.29.0 makes Model Runner V2 the default serving path for all models. The release also adds or expands prompt embeddings, KV-cache behavior, speculative decoding, and support for newer model architectures. A serving upgrade can therefore alter scheduling, memory, and decoding behavior even when an application-facing API does not change. Practical implication: re-measure latency, memory use, batching, and failure behavior before treating this upgrade as routine. vLLM v0.29.0 release
OpenAI’s Python SDK v3.11.0 added expiration controls for service-account keys, while Node SDK v7.13.0 added API-key expiration controls. The change makes credential lifetime an explicit integration capability rather than just a local policy. Practical implication: test expiry and rotation in long-running agents now, including the behavior of work that resumes after a credential’s validity window. Python v3.11.0 Node v7.13.0