AI TrackerInfrastructure news,
for builders.
The latest infrastructure news from every AI provider: launches, price changes and retirements, summarized with what they mean for you.
Vercel starts rolling out Device Bound Session Credentials for vercel.com
Malte Ubl of Vercel says the company is starting to roll out Device Bound Session Credentials (DBSC) for vercel.com. He says DBSC makes it impossible to steal session cookies from a device, so info-stealer malware, for example installed through a supply chain attack, could not take a user's vercel.com session.Anthropic releases FlashProve, an open-source prototype for proving which model produced a response
Anthropic published research on provable inference: a cryptographic proof (a zkSNARK) that a response came from a specific committed set of model weights, checkable by a user who does not have the weights. The post says it could help detect a model silently swapped by a routing service or tampered with by an attacker, and could give evaluators and regulators a way to check what is being served.
Vercel Sandboxes with Drives now start up to 2.4x faster
Vercel says sandboxes that use Drives now start up to 2.4x faster. By its figures, median creation time dropped from 1.7 s to 0.7 s and p95 from 2.2 s to 1.6 s. Vercel says no SDK upgrade or configuration change is needed. The details are in Vercel's changelog entry.Claude Messages API adds message threads in public beta, caching the conversation on the server
Anthropic's ClaudeDevs account says message threads are now in public beta on the Messages API. The API keeps the conversation on the server between turns, so a request carries only the new messages and the previous message ID instead of the full history, and Claude starts responding sooner.
Vercel builds now block known malicious npm packages before install scripts run
Vercel says its builds now automatically block known malicious npm packages before their install scripts can run. No action is needed from users. The details are in Vercel's changelog entry.OpenAI describes Pando, a Git-compatible backend built for Codex-scale development
OpenAI's engineers say the load on the company's monorepo grew so fast with Codex (more branches, pushes, merges, fetches and clones from development environments, research jobs and CI) that, from around the end of 2025, fetches and pushes were slow and merges failed, with reliability at times below 90%. In response they built Pando, a new backend for Git repositories, which OpenAI says now powers all development in its monorepo.MiniMax lists community quantizations for running M3.1 Flash locally
MiniMax posted a roundup of community-made quantized versions of its 428B-parameter open-weight M3.1 Flash, aimed at people running it on their own hardware. MiniMax says lower-bit weights cut the memory needed to load the model and the weight data moved during decoding, and that making this useful takes careful calibration, efficient low-bit kernels and placing experts across GPU and system RAM.
Cloudflare Durable Objects can now wait to apply code updates until they hibernate
Cloudflare Durable Objects can now defer a code update until an object hibernates, so long-running agent turns can finish even when new versions are deployed often. Cloudflare engineers describe it this way: after a deploy, each object waits for its current work to complete before it switches to the new code. The change is documented in Cloudflare's developer changelog dated October 30. One of the engineers adds that more Durable Objects updates are coming, aimed at running long-running agents.
OpenAI announces Private Intelligence, stronger data controls for businesses
OpenAI announced Private Intelligence, a set of data protections for businesses that use its most capable models on sensitive work. OpenAI says it is building the protections in close collaboration with some of its largest customers.Artificial Analysis launches a Search Index that ranks search APIs for agents
Artificial Analysis launched its Search Index, which measures the performance, cost and latency of search APIs across 1,700 tasks from three search benchmarks: BrowseComp, DeepSearchQA and AA-Omniscience. All rows use the same base model and its open-source agent harness, Stirrup. It says agents are expected to search the web far more than humans do by the end of the year.
Claude API raises the maximum request size from 32 MB to 128 MB
Anthropic's ClaudeDevs account says the maximum request size for Messages and Token Counting on the Claude API went from 32 MB to 128 MB. It says image-heavy requests, such as long computer-use sessions and large PDFs, were hitting 32 MB with plenty of context window left, and that this gives about 4x more room.