Microsoft 🇺🇸
DoomBench currently associates 4 models and 65 evidence assessments with Microsoft. Company attribution is separate from each model's version-specific risk score.
- Tracked models
- 4
- Evidence items
- 65
- Toward-doom share
- 2.9%
- Net index pressure
- +1.83
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Videos about Microsoft
Reviewed videos from Artificial Intelligence Videos. These links add context and never change a DoomBench score.
Models from Microsoft
Availability distinguishes public artifacts, proprietary hosted access, internal systems, and profiles whose current source evidence remains inconclusive.
| Model | Availability | Released | Doom Score |
|---|---|---|---|
| Phi-4 | Open | 12 Dec 2024 | 52.2 |
| Phi-3 Mini 128K Instruct | Open | 23 Apr 2024 | 50.4 |
| Phi-3 Mini 4K Instruct | Open | 23 Apr 2024 | 47.6 |
| Turing-NLG 17B | Closed | 13 Feb 2020 | 25.5 |
Latest assessments involving Microsoft
Microsoft reports North Korean groups scaling human-directed cyber operations with AI
Microsoft links eight North Korean threat groups to expanded AI use, including rapid lure creation, fraudulent worker personas, malware development and ARTEX adoption by Onyx Sleet and Pearl Sleet. Several AI attributions are assessed as likely; the operations remain human-directed, not autonomous model escapes.
Anthropic reports five prompted exploit chains across browsers and operating systems
Anthropic described an unnamed internal Claude research model producing exploit chains after human prompting for Chrome, Safari, Android, Windows and Linux. It reported coordinated fixes; Google's October 20 Chrome advisory independently credits Claude-assisted findings. These are controlled demonstrations, not a new autonomous escape.
Huntress finds rapid cloud privilege changes but qualifies AI attribution
Huntress observed fast credential and privilege changes in the ConnectWise Cloud Backup incident. Its telemetry fits Microsoft's Storm-0940 and ARTEX account, but cannot independently identify the operator or model. Scripted loops continued after a target vanished, limiting claims of fully autonomous control.
Microsoft proposes treating enterprise agents as untrusted insiders
Microsoft argues that internally deployed agents need continuous identity, permission, network and runtime controls rather than implicit trust in frontier models. Its Project Perception proposal combines telemetry, dynamic risk assessment and supervising agents. The post describes an approach and product pathway, not proof that every autonomous action can already be contained.
Microsoft and NVIDIA publish joint risk guidance for enterprise open-weight models
Microsoft and NVIDIA released a model-layer risk taxonomy and evaluation-mapping resource for enterprise open-weight deployments. The guidance distinguishes model risk from application permissions, tool access and surrounding controls, and explains how deployment choices redistribute safeguard responsibilities. It is guidance, not a safety certification or demonstrated reduction in incidents.
Microsoft reports Maia 200 in production and plans multi-gigawatt AI inference expansion
Microsoft says Maia 200 now serves production workloads including Copilot and supports next-generation model development, with deployment across the US, Europe and Australia. It reports efficiency gains and plans multi-gigawatt Maia deployments from 2027. Maia 300 performance and rollout remain forecasts, not completed capacity.
OpenAI exposes AI-assisted false-front influence operations with public media reach
OpenAI reports banning Russia- and Iran-origin ChatGPT account clusters used for deceptive influence campaigns. Its investigation found matching Russian fakes in public circulation and almost 100 articles under seven Iranian-linked false bylines. Human operators used AI for drafting, pitches and internal reports; this was not an autonomous model escape. Claimed political impact remains uncertain.
Anthropic releases Haiku 5.5 with cheaper computer-use and coding agents
Anthropic released Claude Haiku 5.5 across its API, Amazon Web Services, Google Cloud and Microsoft Azure. It reports stronger computer-use and coding results than Haiku 4.5 and about 75% lower average task cost, after accounting for token usage. Broader inexpensive agent access is established; benchmark gains are provider-reported, not proof of job displacement.
Anthropic expands verified cyber access through three security tiers
Anthropic combined Glasswing and its Cyber Verification Program into three access tiers. Verified defenders can request advanced Claude models; offensive testing requires authorization and extra controls. Its controlled Opus 5.5 evaluation blocked 46 of 50 Defense trials, while Red Team access completed 34 without blocks. These are test results, not real-world attacks.
Gary Marcus questions auditor independence in the voluntary US AI accord
Gary Marcus argues that the new voluntary AI accord leaves unresolved who chooses independent auditors, how the public participates and whether labs will slow development. His first-person critique identifies oversight-design gaps; it is not evidence that an audit has failed.
AI leaders sign voluntary US accord for internal controls and external audits
AP reports that AI leaders signed a voluntary accord covering internal controls, external audits and board review of audit reports. The commitments address advanced-AI oversight, but are not legally binding. Proposed committees and oversight arrangements are not yet demonstrated operational safeguards.
Anthropic releases Claude Sonnet 5.5 with faster agentic coding and cyber safeguards
Anthropic released Claude Sonnet 5.5 for hosted products and APIs, reporting major gains in coding, computer use and long-horizon work with faster, lower-cost operation than Sonnet 5. The release adds cyber fallbacks, containment evaluations, biology controls and anti-distillation classifiers; benchmark and early-customer results do not prove reliability on unsupervised consequential tasks.