OpenAI expanded Zero Data Retention options for API users to include frontier models while introducing Private Safety Processing.
Teams handling sensitive user data or strict regulatory requirements can send requests to flagship OpenAI models without API inputs being stored on OpenAI servers.
Cloudflare published security research detailing new Spectre attack gadgets and high-resolution remote timers tested against V8 isolation in Cloudflare Workers.
Developers relying on multi-tenant isolate environments like Cloudflare Workers must account for cross-isolate timing leakage when handling sensitive cryptographic material.
Datadog detailed the design of gitretriever, a custom service that handles a 20-fold increase in CI Git traffic while lowering CPU usage.
Infrastructure teams scaling internal developer platforms can adopt Datadog's caching and request-coalescing patterns for heavy Git repository operations.
PlanetScale published an analysis showing how dirty session state in PostgreSQL connections can poison connection pools and corrupt subsequent queries.
Node and Python developers configuring Postgres connection pools should explicitly reset transaction state to prevent cross-request session leakage.
Armin Ronacher explained that changing the reasoning effort parameter in LLM calls invalidates KV caches and forces complete re-computation of prompt tokens.
AI application developers should lock reasoning effort parameters across multi-turn agent turns to preserve cache reuse and keep latency low.
Kent C. Dodds highlighted insights from Shaundai Person emphasizing that cross-team engineering requests require incremental persuasion steps rather than single large demands.
Technical leads pushing for cross-repository architecture changes can improve adoption by breaking proposals into smaller staged conversations.
The most damaging phrase in the language is: we have always done it this way.Grace Hopper
OpenAI Expands Zero Data Retention to Frontier Models
