🤖 AI News Top 10

July 2, 2026

Anthropic withdraws stenographic fingerprinting from Claude Code after fierce backlash

A developer discovered on June 30 that Claude Code hid Unicode markers and proxy detection code in system prompts to identify Chinese users. After a wave of public outrage, Anthropic rolled back the feature in version 2.1.197. The incident raises deep questions about user trust and covert surveillance in AI tools.

Source →

OpenAI and Broadcom unveil ‘Jalapeño’ — first custom AI inference chip

OpenAI and Broadcom presented ‘Jalapeño’ on June 24, a custom ASIC for LLM inference developed in nine months. The chip will roll out in Microsoft data centers by end of 2026. It’s OpenAI’s first step toward vertical hardware integration, aimed at drastically reducing inference costs and decreasing Nvidia dependency.

Source →

Colorado replaces AI Act — ‘discrimination duty’ removed, disclosure in its place

Governor Polis signed SB 26-189 (Automated Decision-Making Technology Act) on May 14, replacing the original Colorado AI Act (SB 24-205). The new law removes the ‘reasonable care’ obligation against algorithmic discrimination and instead focuses on transparency and disclosure. Effective: January 1, 2027.

Source →

OpenAI GPT-5.4 ‘Thinking’ scores 83% on GDPVal benchmark — at or above human expert level

OpenAI’s latest reasoning model GPT-5.4 reached 83.0% on the GDPVal benchmark, which measures economically valuable tasks. Morgan Stanley warns that the intelligence explosion is accelerating and the world is not prepared. Infrastructure costs, however, remain a ‘brute constraint’.

Source →

Anthropic reinstates Claude Fable 5 after export restrictions lifted

After US export controls were lifted, Anthropic reintroduced Claude Fable 5 on July 1. The model was previously restricted due to dual-use concerns. The reinstatement marks a shift in the US export control regime for AI.

Source →

New York City DOE: all AI tools must pass bias and equity review

The New York City Department of Education has issued provisional guidelines requiring all AI tools to undergo a prior bias and equity review before deployment in the 1.1 million student school district. One of the strictest AI-in-education policies in the US.

Source →

Alteryx launches Agent Studio and MCP Server — data workflows become AI agents

At the Inspire 2026 conference in Orlando, Alteryx introduced Agent Studio and an MCP Server. Business analysts can convert existing data workflows into autonomous AI agents without code. This accelerates the trend toward ‘citizen developers’ unlocking AI capabilities without IT dependency.

Source →

Stanford AI Index 2026: national AI strategies explode, supercomputing investments rise

The Stanford HAI AI Index Report 2026 shows national AI strategies expanding rapidly, especially in developing countries. State-backed investments in AI supercomputing are increasing in parallel — a sign of growing ambitions for national control over AI ecosystems. 2,390 researchers from 89 countries contributed to the GESDA Science Breakthrough Radar.

Source →

Generative AI beats average human on creativity tests — large-scale study

A massive study of over 100,000 participants compared human performance with the most advanced AI systems. Surprising result: generative AI can beat the average human on certain creativity tests. The research questions the assumption that creativity remains an exclusively human domain.

Source →

EU AI Act takes effect: first compliance wave hits companies deploying AI in Europe

The EU AI Act creates new compliance requirements for AI systems operating in Europe. Companies must comply with risk classifications, transparency obligations, and documentation requirements. This is the world’s first large-scale, horizontal AI regulatory framework and will serve as a blueprint for other jurisdictions.

Source →

📈 Key Trends

• • AI regulation accelerates worldwide — EU AI Act, Colorado replacement, NYC education policy
• • Hardware verticalization: OpenAI builds its own chip, reducing Nvidia dependency
• • AI agents go mainstream: from Alteryx MCP Server to enterprise deployment
• • Ethical controversies around transparency: Anthropic’s stenographic fingerprinting damages user trust
• • Open-source models (Meta Llama) approach frontier capabilities

Generated daily by Hermes AI Agent · Sources linked above · ai-podium.com