1. OpenClaw CVE-2026-33579: Privilege escalation vulnerability patched
OpenClaw developers have released security patches for a privilege escalation vulnerability (CVE-2026-33579) rated up to 9.8 in impact. The flaw allows any user with basic pairing privileges to silently approve device pairing requests for administrative scope. Once approved, the attacker gains full administrative control over the OpenClaw instance without requiring secondary exploits or user interaction. A compromised administrative device can read connected data sources, exfiltrate credentials, and execute arbitrary tool calls.
2. Microsoft MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 models released on Foundry
Microsoft has launched three new in-house foundational AI models for speech transcription, voice generation, and image creation. The models are available immediately through Microsoft Foundry and a new MAI Playground. MAI-Transcribe-1 supports 25 languages, achieves a 3.0% word error rate on the Artificial Analysis leaderboard, and processes audio at 69x real-time. Pricing for the transcription model starts at $0.36 per hour.
3. Anthropic Claude subscriptions no longer cover OpenClaw usage
Starting April 4th, Anthropic will no longer allow users to apply their standard Claude subscription limits toward third-party harnesses like OpenClaw. Developers using these tools with a Claude login must now purchase separate pay-as-you-go extra usage bundles or use a standard Claude API key. Anthropic is offering a one-time credit equal to the monthly plan cost and discounts on pre-purchased usage bundles to ease the transition. The policy change aims to manage infrastructure demand generated by autonomous AI agents.
4. Google Gemini API adds Flex and Priority Inference tiers
Google has introduced two new service tiers to the Gemini API to provide granular control over cost and reliability. The Flex Inference tier is cost-optimized for latency-tolerant workloads, eliminating the need for batch processing overhead. The Priority Inference tier guarantees that critical traffic will not be preempted during peak platform usage, available at a premium price point. These additions allow developers to balance economic efficiency with performance requirements without managing asynchronous jobs.
5. OpenAI Codex introduces pay-as-you-go pricing for teams
OpenAI has transitioned Codex-only seats within ChatGPT Business and Enterprise plans to a pay-as-you-go pricing model. Teams can now access Codex without paying fixed monthly seat fees, with billing based entirely on token consumption. Additionally, the annual price for standard ChatGPT Business seats has been reduced from $25 to $20 per user. This change lowers the entry cost for teams and simplifies tracking usage expenses.
6. Sarvam 105B and 30B open-weights models released
Sarvam AI has released Sarvam 105B and Sarvam 30B, two new open-weights Mixture-of-Experts models pre-trained from scratch. Both models support reasoning and non-reasoning modes, with the 105B model featuring a 128K context window and the 30B model offering a 65K context window. The models are available under the Apache 2.0 license via Hugging Face and Sarvam's first-party API. While they show strength in select agentic tasks, benchmarks indicate they currently trail leading open-weights peers in overall reasoning and exhibit higher hallucination rates.
7. Qwen3.6-Plus multimodal model announced (Preview)
The Qwen team has announced Qwen3.6-Plus, a new model designed for native multimodal agent workflows. The model features improved environmental perception and sharper multimodal reasoning capabilities compared to previous versions. The Qwen team plans to release smaller-scale, open-source variants of the model in the coming days.