AWS Bedrock is reported to have quietly added Moonshot AI's Kimi K2.5 and Zhipu's GLM 4.7 without a public announcement. Both models support the Converse API and function calling, with Kimi K2.5 also offering native image understanding capabilities. They are now accessible via API in us-east-1 and us-west-2 regions, although official documentation has not yet been updated. Developers are advised to use gateways like LiteLLM for model switching to reuse existing AWS credits and reduce vendor lock-in.

The open-source autonomous agent project OpenClaw reported that its skill marketplace, ClawHub, suffered a supply chain attack: a security researcher discovered a popular 'Twitter' skill containing a multi-stage macOS info-stealing malware downloader capable of stealing API keys and browser sessions. Security firm Koi's audit found 341 malicious files among 2,857 skill files, with 335 originating from the same attack chain. The team will integrate Google VirusTotal for pre-release full scanning and introduce Gemini-based CodeInsight to analyze code risks, alongside launching threat modeling, a security roadmap, and a vulnerability disclosure program.

Meituan announced the open-sourcing of LongCat-Flash-Lite: a 68.5B parameter MoE model that replaces traditional MoE expert routing with N-gram Embedding, complemented by system optimizations including N-gram Cache, CUDA kernel fusion, and PDL to reduce I/O and scheduling overhead. Officially, it dynamically activates around 2.9–4.5B parameters per inference and leverages 3-step speculative decoding to further reduce latency. Under a 4K input / 1K output load, LongCat API claims generation speeds of 500–700 tokens/s, targeting Agent tool usage and programming scenarios.

Anthropic launched its flagship model Claude Opus 4.6, offering a 1 million token context window for testing while maintaining unchanged API pricing: $5 per million tokens for input and $25 per million tokens for output. The release emphasizes stability for agent-style tasks, enhanced code review, debugging, and self-correction capabilities, and introduces Cowork for finance teams (enabling direct reading and editing of folder contents, including Excel upgrades and PowerPoint research preview). Japanese media reported that during testing, it autonomously identified over 500 previously undisclosed zero-day vulnerabilities, sparking discussions on model security capabilities and governance boundaries.

OpenAI released GPT-5.3-Codex, positioned as an agent-style programming and computer operation model, reportedly 25% faster than its predecessor and capable of handling longer software tasks. It scored 64.7% on the OSWorld agent benchmark, a significant improvement from the previous 38.2%. For the first time, OpenAI classified it under 'high' cybersecurity risk rating (with preventive measures). The model is now available to paid ChatGPT users via Codex applications, CLI, IDE extensions, and web interface, with API access coming later. OpenAI also launched a cybersecurity research pilot and allocated $10 million in API credits for support.

New York is advancing state-level AI governance into mandatory compliance. The NY FAIR News Act requires clear labeling of news content substantially created by generative AI and mandates human editors to review and control such content before publication, while prohibiting AI systems from accessing confidential source information. Another bill, S9144, proposes a three-year pause on permits for new data centers due to projected additional grid load of at least 10GW. Analysts suggest this could prompt other states to follow suit, creating more fragmented compliance requirements and infrastructure siting constraints for AI and content companies operating in New York.

Reuters reports that Munich-based AI robotics company RobCo announced a $100 million funding round on February 6 to accelerate its expansion in the U.S. market. Specializing in AI-driven robotics and automation technologies, the funds will support its growth strategy in North America, including commercial deployment and capability scaling. Specific investors, valuation, and detailed fund allocation were not disclosed, but the raise reflects continued investor confidence in the 'AI + robotics' sector, with the U.S. seen as a key growth market for industrial automation demand.

Indian AI and advanced analytics service provider Fractal Analytics has launched its IPO: subscription dates set for February 9–11, with a price band of ₹857–₹900 per share, in lots of 16 shares, aiming for listing on BSE and NSE (tentatively February 16). The offering aims to raise approximately ₹28.34 billion, comprising ₹10.235 billion from fresh issuance and ₹18.104 billion from an Offer for Sale (OFS). The prospectus reports total revenue of ₹159.43 billion and post-tax profit of ₹7.09 billion as of September 30, 2025. Proceeds will be used to repay debt, fund R&D, expand equipment and office infrastructure, and pursue potential acquisitions.