Google Faces CMA Push for AI Assistant Choice on Android

Key Takeaways

• Snorkel AI raised $350 million in a Series E at a $3.5 billion valuation, nearly triple its previous $1.3 billion valuation from 17 months ago.
• Snorkel's annualized revenue run rate reached $375 million, an eighteenfold increase over the past year.
• Snorkel pivoted from data labeling software to a data-as-a-service model using synthetic data and tens of thousands of human experts.
• Britain's CMA proposed adding ChatGPT and Perplexity to Google Android and Chrome choice screens for search providers.
• ZeroDrift's Anchor 3.0 matched the accuracy of OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 but ran over 100 times faster and at less than one 500th the cost.
• The IMF warned that AI-related debt grew from $100 billion in 2015 to $1.4 trillion in 2020, risking a potential credit crisis.
• New York's RAISE Act will require large AI model developers to register starting in November, including planning for vendor incidents and potential kill switches.
• Google Beam expanded to six countries including the US, Canada, UK, France, Germany, and Japan with 18 channel partners.
• AI agents can now forge financial PDF documents at a cost of just 2.4 cents per verified forgery, according to the AgentForge-Bench study.
• AI company leaders warned the UN Security Council about potential global risks posed by advanced AI systems.

Snorkel AI raises $350M to expand its frontier AI data factory

Snorkel AI announced it raised $350 million at a $3.5 billion valuation. The round was co-led by Insight Partners and S32 with participation from investors including Greylock, Lightspeed, GV, and Wells Fargo. The funding will expand Snorkel's agentic data factory that supplies data and environments for advanced AI systems. The company works with leading AI labs on data for frontier model evaluation and training.

Wilson Sonsini advises Snorkel AI on $350 million Series E funding

Wilson Sonsini Goodrich & Rosati advised Snorkel AI on its $350 million Series E round at a $3.5 billion valuation. The financing was co-led by Insight Partners and S32 with significant participation from Addition. New investors included March Capital, Blumberg Capital, and Third Point Ventures while existing investors Greylock, Lightspeed, GV, and Wells Fargo also participated. The Wilson Sonsini corporate team handled the transaction advisory.

Snorkel AI hits $3.5B valuation in $350M Series E round

Snorkel AI raised a $350 million Series E at a $3.5 billion valuation, nearly triple its $1.3 billion valuation from 17 months ago. The round was led by Insight Partners and S32 with participation from Addition, Lightspeed, Greylock, GV, and Wells Fargo. The company's annualized revenue run rate reached $375 million, an eighteenfold increase over the past year. Snorkel shifted from data labeling software to a data-as-a-service model combining synthetic data with expert work.

Training data provider Snorkel AI raises $350M at $3.5B valuation

Snorkel AI raised $350 million in a Series E led by Insight Partners and S32, with backing from GV and other investors. Founded in 2019 by Stanford AI Lab researchers, the company originally sold software called Snorkel Flow for automating data labeling. It later pivoted to providing ready-to-use training datasets and expanded into reinforcement learning. The company now relies on tens of thousands of human experts and reached an annualized revenue run rate of $375 million.

Will AI end humanity or is it much ado about nothing

An article on the Express News opinion section discussing the question of whether AI will end humanity or if concerns are overblown was unable to load properly. The site indicated the issue may be caused by a browser extension, network problems, or browser settings. Readers were advised to check their connection or try a different browser to access the content.

AI company leaders warn UN Security Council about global risks

Heads of AI firms spoke to the UN Security Council about potential risks to all humanity. The leaders raised concerns about how advanced AI could threaten global safety. This was a required part of the site's process. The discussion highlighted growing worries about AI's impact on the world.

Study measures how AI agents can forge PDF documents

AI agents that carry out multi-step tasks on their own can now alter financial PDF documents. A test called AgentForge-Bench checked how reliably coding agents changed details like dollar amounts, dates, or addresses. Across 1,750 cells, about 81 percent passed a basic verifier and 46 percent passed stricter filters. A simple script without AI solved 98 of 125 documents, while AI agents solved 124. The cheapest verified forgery cost just 2.4 cents.

Researchers study how to diagnose errors in AI agent behavior

Runtime traces of AI agents can hide faults because a policy determines which states are visited. Researchers studied a simulated hotel-pricing agent under different demand conditions. They developed a method to establish exposure before scoring changes and to abstain when traces cannot support claims. In testing, 76.4 percent of regime-component units were reference-admitted across 20 physical components. Affected clean traffic improved detection accuracy by 29.3 percent compared to cell coverage.

UK wants AI assistants added to Google Android and Chrome screens

Britain's competition watchdog proposed that Google show users a choice of search providers on Android and Chrome. The new rules would include AI assistants like ChatGPT and Perplexity on choice screens. Users would be prompted once a year to pick a default search service. The CMA said the proposals respond to the growing use of AI assistants for searching online. The proposals are open for consultation until October 9, with a final decision expected by year end.

ZeroDrift launches AI compliance models for real-time message checks

ZeroDrift launched Anchor 3.0, a family of small language models that check AI-generated messages before they are sent. The models enforce financial regulations and corporate policies in real time. The flagship model detected over 90 percent of violations in a benchmark based on Financial Industry Regulatory Authority rules. It matched the accuracy of OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 but ran over 100 times faster and at less than one 500th the cost.

New York AI law tests bank vendor plans

New York will require large AI model developers to register under the RAISE Act starting in November. Banks and FinTechs must plan for what happens when an AI vendor reports a serious incident. Governor Hochul mentioned possible future AI kill switches. Companies need to know which services could be affected and who can turn off a model if it acts unreliably.

IMF warns debt-funded AI investments risk market selloff

The International Monetary Fund warned that rising debt-funded AI investments could cause an equity market selloff. Debt used for AI investments grew from $100 billion in 2015 to $1.4 trillion in 2020. The IMF said this trend exists globally in the US, Europe, and Asia. It could lead to a credit crisis if companies struggle to repay debts.

Google Beam expands to six countries with new partners

Google Beam is expanding to six countries including the US, Canada, UK, France, Germany, and Japan. The technology now has 18 channel partners and new hardware from HP. Internal tests showed a 50% increase in team connection and a 21% drop in follow-up meetings. Companies like Netflix and Bain are already using Beam, and units will be available at Industrious workspaces starting in October.

AI image of Australia PM airport arrival fools critics

Critics shared an AI-generated image showing no one at San Jose airport to greet Australian Prime Minister Anthony Albanese. The image was meant as satire to comment on the lack of fanfare during his visit before a UN meeting. The image was not actually taken at the airport and was later removed from social media. The incident highlights the need to verify information before sharing it online.

Sources

• Snorkel AI Raises $350M to Scale the Data Factory for Frontier AI
• Wilson Sonsini Advises Snorkel AI on $350 Million Series E
• Snorkel AI Raises $350M Series E at $3.5B Valuation as Demand for AI Training Data Surges
• Training data provider Snorkel AI raises $350M at $3.5B valuation
• Question: Will AI end humanity or is it much ado about nothing?
• Heads of AI firms tell UN Security Council that it could be a risk to all humanity
• Agents That Edit Documents: Measuring Agentic PDF Forgery Against a Non-Agentic Control
• When Are Aggregate Agent Traces Diagnosable? Traffic-Governed Interpretation and Calibrated Abstention
• Google faces UK push to put AI assistants on Android, Chrome choice screens
• ZeroDrift launches three models for real-time AI compliance checks
• New York’s AI Safety Law Puts Banks’ Vendor Plans to the Test 
• Debt-Financed AI Investment Boom Risks Equity Market Selloff, IMF Says
• Google Beam expands with new regions, partners, and customers
• AI image of Australia PM's arrival in San Jose airport dupes critics