Satya Nadella urged Microsoft to redesign AI safety by stepping back to reassess the trust architecture rather than treating advanced models as simple black boxes. He called for an emergency-brake capability in which an authorized person can pause or shut down a model mid-task. This shifts systems toward separating the model from its orchestration, adding external safeguards, and requiring documented, tamper-proof evidence for meaningful model actions.
Microsoft launched Microsoft-Decision-1 in Foundry, using a decision model trained on Alibaba’s Qwen3.5-9B instead of OpenAI’s Decisions API model, and positioned it for faster decision-making in applications and agent workflows. It charges $0.042 per million input tokens (with free output). This shifts more decision-model demand toward Microsoft’s Foundry ecosystem while leaving remaining uncertainty about image support, System One compatibility, and how the model’s calibration holds under adversarial input.
Microsoft released Microsoft-Decision-1, a decision-scoring AI model for routing, classification, verification, and agent control that returns calibrated probabilities over fixed answer options rather than generated text. It scored 83.5% average accuracy across Microsoft’s 36 benchmarks with 85 ms p50 latency. The release adds a closed, hosted API that is deployable via Microsoft Foundry or OpenRouter and changes output from prose to JSON option probabilities with pricing at $0.042 per million input tokens and free output.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
