Find published research by company, product, platform or technology.
Showing 21–40 of 188 dossiers
GitHub's multi-model coding orchestration is now GA, with admin-controlled model pools and per-model charges. Local routing remains experimental and cloud-dependent.
GitHub Apps can now authenticate Copilot SDK workloads without a user's credentials. Organisation admins control installation permissions, usage attribution and budgets.
A six-month real-world comparison shows AI code-review agents and specialised fuzzers find substantially different V8 vulnerabilities. Replacing fuzzing with agents would leave gaps, and cost per bug remains unknown.
MLflow CrewAI autologging silently logged API keys; 3.17.0 patches it, but previously exported traces and keys still need remediation.
Cloudflare Computer 0.7 makes agent code resumable and interruptible. The important catch: recovery replays code, and unrecorded external actions can be repeated even if they already succeeded.
Copilot now offers model-reviewed action approvals between manual and Allow All. Unlike OS sandboxing, this probabilistic gate can approve unsafe operations; unattended CLI flags may leave unexpectedly broad permissions.
Cloudflare discloses cross-tenant GLM-5.3 Workers AI responses caused by a KV-cache rejection bug; SGLang v0.5.22 is recommended for disaggregated inference operators.
TorchNeuron changes the software portability equation for Trainium, making it easier to test non-NVIDIA AI accelerators without rewriting the model stack. Beta gaps still demand model-by-model verification.
Jev, CLM and GLiNER2.5-Decide made bounded software decisions look like a distinct model category. OpenAI is now validating the same architectural split with a Luna-powered API designed to answer finite questions rather than generate open-ended prose.
The PyTorch 2.15 RC changes what Apple-silicon developers can run locally: FP8, faster MPS decoding, new GPU operators and a video VAE memory fix. The reported 5–6.5x gain is not universal.
Kimi K3.1 Pro ships via API at $2 input/$10 output per million tokens, with a 2–3x faster Highspeed tier. Independent Artificial Analysis testing scores 55 on its Intelligence Index and 67% on Terminal-Bench 4.0. Open weights remain promised, not released.
The useful shift is architectural: agent permissions no longer have to depend only on the model or harness behaving correctly. OpenShell puts policy enforcement in the execution environment, while Sentry is designed to keep watching from a separate hardware trust domain.
Agent skills join prompts and models as managed runtime configurations; approvals and pinned versions make changing agent behaviour auditable rather than an untracked edit to skill files.
The concrete API price and access limits close a key gap in the October launch: developers can cost Wonder through Model API but must not confuse maximum reasoning effort with Wonder Full.
The unusual risk is that Python startup hooks execute without an application importing LiteLLM, while the public source tag lacked the malicious files. Check the exact installed wheel or image digest, not just the repository.
Google Cloud customers gain a new route to OpenAI models with unified governance and cloud-commitment billing, but broad availability and some privacy features are still forthcoming.
WebMCP is no longer a Chrome-only browser experiment: Microsoft Edge now has its own active origin trial, while ChatGPT’s built-in browser and WordPress Playground show agent-client and platform implementation paths.
Clairic joins Clef as a specialised one-pass model on Workers AI; typed extraction has a different failure model from LLM-generated JSON, with vendor-only benchmarks and preview limitations.
A detailed engineering case study links a 68% AI-review cost reduction to prompt-cache structure and effort tuning, with security-review trade-offs still unresolved.
WebKit's Safari MCP server is available in stable Safari 27.0 as well as Technology Preview. It runs locally, but captured page data passes to the connected agent; remote automation must be enabled explicitly.