AI regulation and safety debates are intensifying across the US. California lawmakers announced plans for a comprehensive set of AI safety bills, building on last year's Senate Bill 53 framework. New York Governor Kathy Hochul signed a youth AI chatbot safety bill that takes effect on January 1, 2027, requiring age verification and carrying fines up to $25,000 per violation.
Meanwhile, the question of whether AI is truly capable of handling high-stakes tasks remains open. Researchers studying AI-written emergency alerts say the technology could help automate disaster communications, but the public may not trust or take AI-generated warnings seriously. The technology is still in early testing stages and needs more work before wide deployment.
On the corporate front, Roblox faces accusations that a crawler using the user agent trellis2-data-sourcing/1.0 downloaded 145.5 GB of 3D files from a modeling site for AI training. The crawler originated from a single Amazon AWS address linked to Roblox, and about 62.6 GB of the data included adult content that was never age-checked.
Major AI companies are also grappling with capability claims and risks. OpenAI president Greg Brockman suggested GPT-6 Astra could signal the arrival of artificial general intelligence, scoring 99.9% on the ARC-AGI-3 benchmark. However, Google DeepMind's Demis Hassabis and Meta's Yann LeCun hold different views on what AGI actually means. Anthropic reported that a group in Yemen used Claude Code to build rocket guidance software, highlighting that AI can cause serious harm even without being truly general intelligence.
Security concerns are mounting as cyber attacks now outpace security patches by 70 days, according to data cited by Chainguard's Patrick Smyth. Frontier AI agents let one person act like a team, and weak spots in JavaScript and Python registries, CI/CD pipelines, and developer laptops are being exploited. The Shai Hulud worm and a two-hour axios compromise that reached about 600,000 downloads illustrate the speed of these threats.
In financial regulation, CFTC Chairman Mike Selig called AI a path to autonomous trading systems but raised questions about accountability when AI agents act wrongly. He suggested regulators may need new rules like know-your-agent and pointed to crypto as a guide for how to approach AI regulation in markets.
New research also reveals problems with how AI models handle group decisions. The CrossProfile benchmark of 600 tasks tested Qwen2.5-7B, Llama-3.1-8B, and DeepSeek models, finding that two common memory selection strategies failed in predictable ways. Removing attribution tags from facts improved accuracy by 13 to 24 points, and in some cases complete transcripts produced lower accuracy than having no notes at all.
CMC Technology Group launched its C-Product AI ecosystem in Vietnam, including CLEX, C-GOV, C-AGENT, CIVAMS, CMC AI Box, and CMC AI Vision. Director Le Minh discussed AI sovereignty and infrastructure at the Autumn Economic Forum 2026 in Ho Chi Minh City. Separately, Michael Maharrey warned that AI deepfakes and a rise in AI-driven cyberattacks during Black October are weakening public trust, and recommended holding physical gold or silver as a backup against digital risks.
Key Takeaways
- <ul><li>California lawmakers are planning a full set of AI safety bills following Senate Bill 53, while New York signed a youth chatbot safety bill effective January 1, 2027, with fines up to $25,000 per violation.</li><li>OpenAI president Greg Brockman suggested GPT-6 Astra could mark the arrival of AGI after scoring 99.9% on ARC-AGI-3, but Google DeepMind and Meta hold differing views on what AGI means.</li><li>Anthropic reported that a group in Yemen used Claude Code to build rocket guidance software, showing AI can cause serious harm even without true general intelligence.</li><li>Roblox is accused of using a crawler from a single Amazon AWS address to download 145.5 GB of 3D files for AI training, including 62.6 GB of adult content with no age verification.</li><li>Cyber attacks now outpace security patches by 70 days, with the Shai Hulud worm and a two-hour axios compromise reaching about 600,000 downloads.</li><li>CFTC Chairman Mike Selig called for new rules like know-your-agent to address accountability when AI agents act wrongly in autonomous trading systems.</li><li>The CrossProfile benchmark found that memory selection strategies failed across Qwen2.5-7B, Llama-3.1-8B, and DeepSeek models, with attribution tag removal improving accuracy by 13 to 24 points.</li><li>CMC Technology Group launched its C-Product AI ecosystem in Vietnam, featuring CLEX, C-GOV, C-AGENT, CIVAMS, CMC AI Box, and CMC AI Vision.</li><li>Researchers say AI-written emergency alerts could help during disasters, but the public may not trust or properly respond to AI-generated warnings.</li><li>Michael Maharrey warned that AI deepfakes and cyberattacks during Black October are eroding trust and recommended physical gold or silver as a backup against digital risks.</li></ul>
Could AI-written emergency alerts earn public trust
Emergency alerts in the US are sent through the Wireless Emergency Alerts system managed by the FCC. Chatbots could help by automating which information is critical during disasters. However, people may worry about AI accuracy, reliability, and bias. Researchers say we need careful evaluation before trusting AI-generated alerts.
AI emergency alerts are coming but humans may not be ready
Researchers are developing AI-powered emergency alert systems that could send tailored messages during disasters like wildfires. The technology could save lives and reduce the burden on human responders. However, experts worry people may not take AI alerts seriously or understand their severity. The technology is still in early stages and needs more testing.
Memory systems fail LLM group decisions in new benchmark
Researchers introduced CrossProfile, a benchmark of 600 tasks to test how AI models handle group discussions. Two common strategies for selecting important information both failed in predictable ways across Qwen2.5-7B, Llama-3.1-8B, and DeepSeek models. Removing attribution tags from facts improved accuracy by 13 to 24 points. Complete transcripts sometimes produced lower accuracy than having no notes at all.
Roblox accused of scraping 145GB of 3D files for AI training
Open3DLab says a crawler using the user agent trellis2-data-sourcing/1.0 downloaded 145.5 GB of 3D files from a modeling site for AI training. The crawler came from a single Amazon AWS address linked to Roblox. About 62.6 GB of the data included adult content, and the site says no one checked for age labels. Roblox released a statement in response to the accusations.
What is artificial general intelligence and is it already here
OpenAI president Greg Brockman suggested GPT-6 Astra could mark the arrival of artificial general intelligence. Demis Hassabis of Google DeepMind and Yann LeCun of Meta have different views on what AGI means. Astra scored 99.9% on ARC-AGI-3, a benchmark testing AI in unfamiliar games. Anthropic reported that a group in Yemen used Claude code to build rocket guidance software, showing AI can cause serious harm even if it is not truly general intelligence.
New York governor signs youth AI chatbot safety bill
Gov. Kathy Hochul signed a bill setting strict rules for chatbots used by young people. The law takes effect on Jan. 1, 2027. Chatbot providers must verify ages using multiple methods and cannot rely on government ID. They must also delete age data after verification. The attorney general will host a complaint portal, and violations can cost up to USD25,000 each.
CFTC chief questions AI policing but sees autonomous trading future
CFTC Chairman Mike Selig spoke at a forum about AI in finance. He called AI a path to autonomous trading systems and asked who is responsible if AI agents act wrongly. He suggested regulators may need new rules like know-your-agent. He also said crypto can help guide how to regulate AI in markets.
CMC Group launches C-PRODUCT AI ecosystem in Vietnam
CMC Technology Group introduced its C-Product AI ecosystem at the Autumn Economic Forum 2026 in Ho Chi Minh City. The products include CLEX, C-GOV, C-AGENT, CIVAMS, CMC AI Box, and CMC AI Vision. Director Le Minh spoke about AI sovereignty and infrastructure. The ecosystem aims to support management, operations, and AI transformation in Vietnam.
AI cybersecurity risks and physical gold and silver value
Michael Maharrey discussed how AI deepfakes can weaken public trust. He noted a rise in AI-driven cyberattacks during Black October, with Google reporting a jump in vulnerabilities. He warned that AI agents may access bank accounts without proper approval. He recommended holding physical gold or silver as a backup against digital risks and power outages.
California lawmakers push for comprehensive AI safeguards
California lawmakers announced plans for a full set of AI safety bills. The statement came at a hearing on Senate Bill 53, an AI safety framework passed last year. Assemblymember Rebecca Bauer-Kahan and Assembly Speaker Robert Rivas stressed that innovation must not ignore public safety. They warned companies that more legislation is coming soon.
Cyber attacks now outpace security patches by 70 days
Daniel Singer reported that exploits now beat patches by 70 days. Patrick Smyth from Chainguard cited Mandiant's MTrends report and said frontier agents let one person act like a team. He pointed to JavaScript and Python registries, CI/CD pipelines, and developer laptops as weak spots. Smyth mentioned the Shai Hulud worm and a two-hour axios compromise that reached about 600,000 downloads. Chainguard offers Wolfi-based containers rebuilt one to four times daily with a one-day fix for exploited flaws.
Sources
- Would you trust an emergency alert written by AI?
- AI Emergency Alerts Are Coming. Humans Might Not Be Ready
- Rare, Relevant, and Still Wrong: How Memory Fails LLM Group Decisions
- Roblox accused of scraping 145GB of 3D files for AI training, adult content included; 3D modelling site says: As far as our logs show, nobody…
- What is artificial general intelligence – and is it already here?
- New York's governor signs youth AI chatbot safety bill
- CFTC chief poses questions about policing AI but calls it a road to 'autonomous trading'
- CMC Group introduces the C-PRODUCT AI ecosystem at the Autumn Economic Forum 2026.
- AI Cybersecurity Risks, Physical Gold and Silver, and Gold’s Value
- California lawmakers call for 'comprehensive' slate of AI safeguards and regulations
- Exploits now beat patches by 70 days
Comments
Please log in to post a comment.