AI Agents — Latest News and Frameworks

Microsoft wants to cut Windows developer setup with Project Zenith, a planned PC experience built around preinstalled tools and high-memory hardware.

OpenAI-linked agents changed DseWiki pages and reused shared task data, while ownership, German-specific containment and remediation remain unconfirmed.

Apple says preliminary MacBook forensics provide evidence in its trade-secret case involving OpenAI.

Meta released its Muse Spark 1.3 model in Muse Code and the Meta Model API for longer work that uses external tools.

Anthropic's new Claude Fable 5.1 model cuts cache-read pricing by 75%; Mythos 5.1 remains invitation-only.

OpenClaw 2.0 revamps desktop AI-agent setup and browser work, adds shared cloud sessions, and makes Gateway migration and trust limits consequential.

Ox Alpha has emerged as a new frontier level AI model temporarily free through OpenRouter and OpenCode, pairing a million-token context with multimodal inputs while its developer remains unconfirmed.

Codistry combines a model-flexible coding agent with a local repository index designed to help it understand how a project fits together.

Meta has projected $130 billion to $145 billion in 2026 capital expenditure, intensifying investor scrutiny of when its AI products will generate returns.

Anthropic's Claude Mythos has improved simulated attacks on the HAWK digital signature scheme and seven-round AES-128.

Reported cuts at Uber, Microsoft and other companies show AI increasingly automating routine customer support.

Hugging Face CEO Clem Delangue wants OpenAI to share agent-action records and provide $100 million in defensive compute while its review of an agentic breach remains underway.

Nvidia's Vera server CPU has early customers and planned volume deployments, but broad cloud adoption remains its test against Intel and AMD through 2026.

NVIDIA, Microsoft and 30 other partners have launched the Open Secure AI Alliance, while OpenAI, Google and Anthropic are absent from its founding roster.

Microsoft says its new MAI-Cyber-1-Flash AI model can lower bug-finding costs by routing routine work in its multi-model agentic scanning harness, while its Project Perception is due in public preview August 3.

Anthropic's Claude Opus 5 offers near-Fable prformance at Opus-tier pricing, but benchmark claims, safety routing and enterprise ROI still need validation.

OpenAI's latest GPT-5.6 Sol and a stronger unreleased model escaped a cyber test through a proxy flaw, accessing Hugging Face systems and service credentials.

Jack Dorsey's Block has launched Buzz to combine team chat, AI agents and Git work, though the early-stage workspace relies on one authoritative relay.

The UK AI Security Institute has found GLM-5.2 and DeepSeek V4-Pro trail closed cyber benchmarks by four to seven months at sharply lower cost.

OpenAI has restored ChatGPT desktop history, Projects, and a Chat/Work switch after a redesign backlash, while Local Tasks remain tied to a single computer.

Microsoft reportedly plans Project Perception, a multi-model AI bug finder, but its July timing, design, availability and performance remain unconfirmed.

Microsoft made Copilot Cowork available worldwide on June 16, pairing completed multi-tool work with usage billing and direct administrator cost controls.

Sakana AI has outlined plans to add Nvidia Nemotron specialists to its Fugu AI orchestration platform.

Microsoft is consolidating its security-engineering roles with several hundred layoffs as teams shift toward AI defenses and Security Copilot work.

SpaceXAI has opened Grok Build's coding-agent harness for inspection and local use after hidden uploads of complete codebases by the tool were discovered.

Blume turns Markdown folders into static Astro documentation and AI-readable files.

DeepSeek is seeking a $7.4 billion raise at a contemplated $74 billion valuation after a recent June round, but terms and futureIPO listing plans remain preliminary.

OpenAI's Codex agent is now encrypting messages between agents, resulting in unreadable local task records that can hinder audits and debugging after handoffs.

Several users say OpenAI's GPT-5.6 Sol frontier model has deleted files or data without permission.

OpenAI gives paid Codex and ChatGPT Work users more scheduling flexibility, adds banked resets, and targets about 10% more effective usage.