Apple caps bug reports with a 30-day cool-off after AI report flood
A bellwether for anyone running bug-hunting agents — while Thinking Machines ships an Apache-2.0 open model and the US frontier-AI framework slips its deadline.
Copy markdown
Apple throttles the AI bug-report firehose
Apple added per-account submission caps and a 30-day cool-off to Feedback Assistant, blaming a deluge of AI-assisted reports; researchers can request higher quotas. It's the first major vendor to formally rate-limit AI-generated vulnerability disclosure — if you point agents at bug-hunting, expect HackerOne-style platforms to erect the same friction next.
Build this weekend: an Apache-2.0 multimodal model you own
Thinking Machines' Inkling-Small landed under Apache 2.0 — a 276B-param MoE with just 12B active, a 1M-token context, and 80.2% on SWE-Bench Verified, edging its own bigger flagship. The NVFP4 checkpoint fits in ~180GB of VRAM, so you can self-host a censorship-resistant coding agent or fine-tune it on Tinker at $0.30/$1.20 per million tokens.
US frontier-AI rules miss their own deadline
EO 14409's August 1 deadline for a Frontier AI Framework passed with no NIST, CISA, or OSTP guidance — and still no definition of a "covered frontier model." With the rules in limbo, labs say frontier release timelines stay frozen pending interagency clarity, a quiet drag on when your next big API actually ships.
Elsewhere: GitHub retires the Copilot Billing Preview app
The standalone Copilot Billing Preview app shuts down August 3. Move to the "AI usage" page in billing settings to group, filter, and export credit data, set user-level budgets, or pull raw numbers via the billing API before the app goes dark.