AI Weekly - Week 50/2025
TL;DR
This week in 30 seconds:
- Open Standard: MCP donated to Linux Foundation – OpenAI, Google, Microsoft, AWS are joining
- GPT-5.2: OpenAI’s response to Code Red with 400K context and 93.2% GPQA Diamond
- Policy Bombshell: Trump signs Executive Order against State AI laws
The Story of the Week
MCP Becomes Universal Standard – and the Entire Industry is Joining In
On December 9, 2025, something historic happened for the AI industry: Anthropic donated the Model Context Protocol (MCP) to the Linux Foundation. [2] But that was just the beginning.
The newly established Agentic AI Foundation (AAIF) unites the industry’s biggest competitors for the first time: Anthropic, OpenAI, Google, Microsoft, AWS, Cloudflare, and Bloomberg as Platinum members. [3] Three projects form the foundation:
- MCP (Anthropic): The universal standard for AI tool integration – over 10,000 published MCP servers ranging from developer tools to Fortune 500 deployments [2]
- AGENTS.md (OpenAI): Already adopted by 60,000+ open-source projects, including Cursor, Devin, GitHub Copilot, Gemini CLI [3]
- Goose (Block): Open-source agent framework for local, trustworthy AI workflows [3]
“We are seeing AI enter a new phase, as conversational systems shift to autonomous agents that can work together.” - Jim Zemlin, Executive Director, Linux Foundation [3]
The timing is no accident: the shift from chat to agents requires interoperability. If your Claude agent needs to talk to a GPT tool and a Gemini service, you need standards. MCP delivers exactly that. [2]
What changes for developers: Nothing dramatic – existing maintainers retain technical control. But now vendor neutrality is guaranteed. MCP belongs to no one and to everyone simultaneously. [2]
Bottom Line: The AI industry has learned: standards win over walled gardens. Anyone building MCP servers now is building for the entire ecosystem, not just one vendor.
Other Top Stories
GPT-5.2: OpenAI’s Response to Code Red
Just 10 days after the internal “Code Red” memo, OpenAI delivered GPT-5.2 on December 11 – along with the specialized GPT-5.2 Codex for software engineering. [1]
Three Variants: [1]
- Instant: Speed-optimized for routine queries
- Thinking: For complex work like coding, long documents, math
- Pro: Maximum accuracy for enterprise-critical tasks
The Numbers: [1]
- 400K context window (2.5x GPT-5.1)
- 93.2% on GPQA Diamond (PhD-level Q&A)
- 80.0% on SWE-bench Verified (just slightly behind Claude Opus 4.5’s 80.9%)
- 100% on AIME 2025 without tools
- 38% fewer errors than predecessor
Pricing: $1.75 input / $14 output per 1M tokens – more expensive than Gemini 3 Flash, but with more capabilities. [1]
Reality Check: The benchmarks are impressive, but this doesn’t solve OpenAI’s growth problem. Gemini continues growing faster, and Claude dominates enterprise coding.
Disney Opens 200+ Characters for Sora
The month’s most surprising partnership: Disney and OpenAI are allowing users to generate videos featuring over 200 Disney characters via Sora – from Mickey Mouse to Marvel heroes. [5]
This is a radical shift for Disney, which has traditionally guarded its IP like treasure. The deal shows: even the most conservative content owners recognize that AI-generated content is inevitable. Better to participate in a controlled way than to be left behind.
Trump’s AI Executive Order: Federal vs. State
On December 11, President Trump signed the Executive Order “Ensuring a National Policy Framework for Artificial Intelligence”. [4] The core message: Federal law should override State laws.
What the Order Actually Does: [4]
- AI Litigation Task Force: To review State AI laws for constitutional compliance within 30 days
- Commerce Department: 90 days to identify “onerous” State laws
- Funding Pressure: States with “problematic” AI laws lose broadband funding
The Exceptions: Child safety, AI infrastructure, and government procurement remain state matters. [4]
Industry Reaction: Mixed. Big Tech welcomes standardization, but 24 State Attorneys General have already filed protests – citing violations of the 10th Amendment and dormant commerce clause arguments. [4] The tension between innovation and consumer protection remains unresolved.
Critical Voices: Consumer advocacy groups (EFF, NCLC) warn that the order could endanger important consumer protection laws in California (privacy) and Massachusetts (transparency).
Quick Hits
In Brief:
- Accenture-Anthropic Multi-Year Deal: Bringing enterprises from AI pilots to production – concrete implementation support instead of just API access. [6] Details
- Gemini 3 Flash Release: Google makes Flash the default in Gemini App + Search. 81.2% MMMU-Pro, 90.4% GPQA Diamond. Launch
- Gemini Deep Research: Reimagined research agent based on Gemini 3 Pro – now with Interactions API for developers. More
- Adobe ChatGPT Integration: Edit directly in ChatGPT: Photoshop, Acrobat, Express via natural language. TechCrunch
- NIST AI Cybersecurity Framework: First official standard for AI-specific security risks (data poisoning, model theft). NIST
- xAI + El Salvador: Grok for 1M+ students across 5,000 public schools – first nationwide AI education program. News
Tool of the Week
Gemini Deep Research - PhD-Level Research on Demand
Google’s newly redesigned research agent is based on Gemini 3 Pro and is now available to developers via the Interactions API. The tool:
- Systematically searches sources (web, papers, docs)
- Synthesizes findings into structured reports
- Delivers PhD-level insights with source attribution
The game changer: you can now embed Deep Research directly in your own apps. Instead of manually writing prompts, you call an API and get research reports back.
Fail of the Week
1,000+ AI Laws in 47 States – and No Overview
Over 1,000 AI-related bills were introduced across US states in 2025. In healthcare alone: 250+ bills in 47 states. [4] The problem: no company can be compliant when every state makes its own rules.
The Trump Executive Order responds to this – but with a sledgehammer instead of consensus. [4] The result: legal uncertainty for years as courts determine constitutionality.
Root Cause: AI regulation became a campaign issue before technical standards existed.
What We Learn: Fragmented regulation is just as harmful as no regulation. The industry should have pushed for self-imposed standards earlier.
Number of the Week
10,000+ MCP Servers
Number of published Model Context Protocol servers – ranging from developer tools to Fortune 500 deployments [2]
A year after launch, MCP has become the de facto standard for AI tool integration. The Linux Foundation donation now secures its future. [2] For comparison: REST APIs took 5+ years to achieve similar adoption.
Reading List
For the Weekend:
- MCP joins the Agentic AI Foundation - The official Anthropic announcement with governance details (5 min)
- Introducing GPT-5.2 - OpenAI’s launch post with benchmarks and capabilities (7 min)
- Trump’s AI Executive Order Analysis - Legal deep dive by Sidley Austin (10 min)
Next Week
What’s Coming:
- Gemini 3 Flash Rollout – more regions and integrations
- NY RAISE Act – Hochul expected to sign for AI transparency
- OpenAI GPT-5.2 Pro – Enterprise rollout begins
- Anthropic Genesis Mission Details – DOE partnership gets concrete
This newsletter was researched and written AI-assisted. Hero image generated with Pollinations.ai.