Google, OpenAI Back Shared AI Training Control

Key Takeaways

• <ul><li>Google and OpenAI will both support a shared robots.txt AI training signal called __usage_ai-training, letting website owners opt out of AI training across multiple providers at once.</li><li>Amazon plans to sell its Trainium AI chips to other companies starting in late 2027, competing with NVIDIA in the AI chip market.</li><li>Amazon's custom chips Graviton and Trainium drove AWS growth with revenue commitments growing over 20 times year over year, saving customers over $2 billion in Q3.</li><li>Epoch AI found that GLM 5.3 from Z.ai can exploit a Firefox patch gap, potentially affecting over 200 million Firefox users by stealing passwords and authentication cookies.</li><li>China published a 22-point AI deployment guide for its shipping industry, targeting maritime chokepoint monitoring, equipment failure prediction, and decarbonisation planning.</li><li>Alphabet reported $51.4 billion in capital spending with negative free cash flow, while CoreWeave shares fell 36% as investors demand cash discipline from AI infrastructure companies.</li><li>AMD and Palantir outperformed in a selective AI stock pullback while Alphabet's stock fell 10% over the past week.</li><li>Wall Street veteran Ed Yardeni argues that spending constraints on AI could actually extend the boom by forcing better spending habits and reducing over-investment risk.</li><li>Verification-Guided Reflow (VGR) improved verified solvability in retrosynthesis planning from 51.1% to 71.1% by incorporating reaction verification into the search process.</li><li>The Center for Civic Futures granted Nava Labs funding to use AI to map legacy government technology systems, aiming to cut the process from months to weeks.</li></ul>

Google adopts new AI training control for websites

Google is adopting a new AI training control being developed by the IETF. It lets website owners set AI training preferences once in robots.txt instead of listing each AI provider separately. Google plans to implement this in 2027. The system uses a shared policy token so publishers can opt out across multiple providers at once.

OpenAI supports common robots.txt signal for AI training

OpenAI announced it will honor a shared signal in robots.txt so websites can express AI training preferences to multiple providers at once. This removes the need for website operators to manage separate rules for each AI company. OpenAI plans to implement the signal in the coming months. Existing robots.txt controls will still apply and be respected.

OpenAI to honor shared robots.txt signal for AI training

OpenAI announced on October 30, 2026 that it will honor a shared robots.txt signal for AI training preferences. The signal uses a policy token called __usage_ai-training and follows the existing robots.txt standard without adding new grammar or directives. OpenAI said the signal does not control crawling behavior and will be implemented in the coming months.

Amazon to sell AI chips to other companies in 2027

Amazon says it will start selling its Trainium AI chips to other companies beginning in late 2027. This move could compete with NVIDIA by offering an alternative for AI chip buyers. Amazon open-sourced TorchNeuron so developers can run PyTorch code on Trainium without rewriting it. The chip sales plan could help Amazon turn its heavy spending into cash.

Amazon says custom chips now drive AWS growth

Amazon says its custom chips Graviton and Trainium are now driving AWS growth. Revenue commitments for each chip family grew more than 20 times year over year. Customers saved over $2 billion in Q3 using the custom silicon. Amazon open-sourced TorchNeuron to help developers run PyTorch code on Trainium without rewriting it.

Epoch AI finds GLM 5.3 can exploit Firefox patch gap

Epoch AI published research on October 29, 2026 showing that GLM 5.3, an open-weight model from Z.ai, can exploit the browser patch gap. The model built a webpage that could read files from a Firefox user's computer without any user interaction. Any Firefox version released between September 17 and October 26, 2026 could be compromised using only information available at the time. The exploit could steal passwords, authentication cookies, and private information from over 200 million Firefox users.

China plans AI deployment across shipping industry

China's transport ministry published a 22-point AI deployment guide addressed to three state-owned giants including China Cosco Shipping and China Merchants Group. The plan aims to embed AI across shipping to monitor maritime chokepoints, streamline operations, and support the green transition. AI tools will be used for predicting equipment failures, detecting unusual vessel movements, and coordinating emergency responses. The guide also covers commercial uses like route design, cargo flow management, and decarbonisation planning. It sets no binding deadlines, leaving priorities to provincial authorities and state-owned groups.

AI infrastructure boom shifts as Alphabet spends $51.4 billion

Alphabet reported $51.4 billion in capital spending and negative free cash flow for the second quarter, while CoreWeave faced debt pressure with shares falling 36% and bond yields near 15.8%. Investors are shifting from rewarding AI infrastructure spending to demanding cash discipline and clear funding milestones. The market reassessment affects how AI infrastructure companies should approach capital allocation and financial planning. Firms now need to demonstrate profitability and establish clear funding milestones to maintain investor confidence.

AI stocks split as compute and software lead while Alphabet falls

The AI stock market is experiencing a selective pullback with compute and data software companies leading while Alphabet struggles. AMD and Palantir are bucking the trend with strong performance while Alphabet's stock price fell 10% over the past week. The rotation reflects a shift in investor sentiment toward companies focused on compute and data software. Alphabet's core business is struggling to keep up with competition, leading to declining investor confidence.

New method improves retrosynthesis with reaction verification

Researchers introduced Verification-Guided Reflow (VGR), a method that incorporates reaction verification into AND/OR search for retrosynthesis planning. On Retro190, feeding verification results into search increased verified solvability from 51.1% to 62.6% compared to route evaluation without feedback. The strategy of early pruning and delayed guidance increased verified solvability from 61.1% to 71.1% using fewer proposal calls and unique verifier queries. This approach helps identify and correct infeasible reactions during the search process rather than after it finishes.

Wall Street veteran says AI slowdown could extend boom

Stocks dropped this week as investors worry about the future of the AI boom after mixed earnings reports. Wall Street expert Ed Yardeni believes that financial, physical, and political limits on AI spending might actually make the boom last longer. He notes that huge companies like SpaceX, Meta, and Microsoft are raising billions in money to fund their projects. Yardeni argues that these constraints will force better spending habits and reduce the risk of over-investment. He also suggests that a slower pace of building data centers could help lower inflation for consumers.

Center for Civic Futures gives grant to Nava Labs for AI tools

The Center for Civic Futures awarded a grant to Nava Labs to help states update old technology systems for public benefits. The project will use artificial intelligence to speed up the process of mapping legacy systems from months down to weeks. Nava Labs plans to work with several states to test these new tools and create a shared roadmap for modernization. This effort aims to improve how quickly and reliably agencies can serve people using Medicaid and the Supplemental Nutrition Assistance Program. The organization will also form an advisory group of state governments to provide feedback on the new approach.

Sources

• Google Adopts New AI Training Control
• Making it easier for websites to express AI training preferences
• OpenAI to Honor Shared Robots.txt Signal for AI Training Preferences
• Amazon Says It Will Start Selling Its AI Chips to Other Companies in 2027
• Amazon Says Custom Chips Now Drive AWS Growth
• Epoch AI Reports Open-Weight GLM 5.3 Exploits Firefox Patch Gap
• China maps state-led push to deploy AI across shipping
• The End of the AI Infrastructure Boom, and What Alphabet's Capital Expenditures and CoreWeave's Setbacks Mean for Valuation and Investment Strategy
• AI Stocks Are Splitting: Compute and Data Software Lead While Alphabet Stumbles
• Verification-Guided Retrosynthesis: Reaction Feedback and Evidence Timing
• An AI slowdown that takes some air out of the exuberance could actually extend the boom, Wall Street veteran says
• Center for Civic Futures awards grant to Nava Labs to help modernize state benefits technology with AI