US security agencies issued an advisory warning that Chinese AI firms may be training models using outputs from US systems, urging a coordinated response. In response, China rejected these claims as groundless, accusing the US of attempting to monopolize computing power and interfere with normal commercial activities. China emphasized that model distillation is a neutral industry practice used globally.
A separate cyber threat exploited two PaperCut software flaws to hack 395 organizations across 48 countries. The attacker, likely speaking Russian, used hundreds of AI agents powered by OpenAI's Codex and a DeepSeek model to breach systems. The campaign moved rapidly, compromising a high school in just seven minutes, with schools in the US and UK hit hardest.
Google announced updates to its Gemini app and search products to include election information from state governments and The Associated Press. However, a study revealed that 29% of AI responses regarding voting were incomplete, inaccurate, or outdated, with Spanish language responses showing even lower accuracy.
In healthcare, California nurses celebrated the passage of bills AB 1979 and AB 2575, which aim to make AI systems transparent and explainable while giving patients appeal rights. Meanwhile, experts note that only 42% of health systems currently use provider data effectively, highlighting the need for better data before advancing AI tools.
Other developments include Akamai launching an AI Assistant for its Web Security Analytics platform to help analysts investigate threats using natural language prompts. Researchers also developed a new AI method called Decision-Focused Active Learning to improve the recovery of critical materials, reducing the number of tests needed from 48 to 16 to 24.
Caterpillar Inc. reported its biggest quarter in history with a 73% revenue increase in its construction industries segment, driven by demand for heavy machinery and autonomous systems supporting the AI buildout. The company is positioning itself as a key player in the AI infrastructure space through its expansion in autonomous technologies.
Key Takeaways
- US security agencies warned of Chinese AI distillation, prompting China to reject the claims and warn of retaliation.
- An attacker used AI agents powered by OpenAI's Codex and DeepSeek to hack 395 organizations via PaperCut vulnerabilities.
- Google added election data to Gemini, but a study found 29% of AI voting responses were inaccurate or outdated.
- California passed bills AB 1979 and AB 2575 to regulate AI transparency and patient rights in healthcare.
- Only 42% of health systems effectively use provider data, creating a bottleneck for AI adoption in the sector.
- Akamai launched an AI Assistant for Web Security Analytics to help analysts investigate threats using natural language.
- New AI methods reduced critical material recovery tests from 48 to 16-24, improving efficiency at Pacific Northwest National Laboratory.
- Caterpillar saw a 73% revenue jump in its construction segment due to rising demand for AI infrastructure machinery.
- The PaperCut attack targeted 280 victims for credentials and gained admin privileges at 12 organizations.
- China accused the US of trying to monopolize computing power while claiming model distillation is a neutral global practice.
US Security Agencies Warn of Chinese AI Distillation
Three US security agencies issued an advisory on Tuesday claiming that Chinese AI firms trained their models using outputs from US models. They called for a coordinated response from the government, private sector, and allies to address this issue. The agencies recommended that US AI companies improve detection of suspicious accounts and use downgraded models to handle suspected distillation requests. They also urged providers to share intelligence across cloud platforms and API aggregators.
China Rejects US AI Distillation Claims and Warns of Retaliation
China rejected US allegations of malicious AI distillation as groundless on Wednesday, warning of retaliation if Washington uses these claims to suppress Chinese AI development. The Ministry of Commerce stated that US accusations of industrial-scale copying lacked factual and legal grounds. Officials accused the US of trying to monopolize computing power and interfere in normal commercial activities. China emphasized that model distillation is a neutral industry practice used by companies worldwide, including American firms.
AI Attack Exploited PaperCut Flaws to Hack 395 Organizations
A threat actor likely speaking Russian used hundreds of AI agents to exploit two PaperCut software flaws and hack 395 organizations across 48 countries. The attack targeted CVE-2026-81578 and CVE-2026-82078 vulnerabilities, compromising at least 440 PaperCut instances. Most victims were in the education sector, with the United States being the most targeted country. Attackers harvested credentials from 280 victims and gained administrator privileges at 12 organizations in some cases.
Hundreds of AI Agents Helped PaperCut Attacker Hit 395 Orgs
An unknown attacker used hundreds of AI agents powered by OpenAI's Codex and a DeepSeek model to exploit PaperCut vulnerabilities and breach 395 organizations. The campaign, traced to an IP address on August 31, moved extremely fast, with one high school compromised in just seven minutes. Although the attacker instructed agents to avoid 28 countries, some organizations in those nations were still hacked. Schools in the US and UK were hit hardest, with 204 and 59 victims respectively.
Voters Turn to AI for Election Help but Answers May Be Wrong
Google announced it would add election information to its Gemini app and search products after previously restricting such content. The update includes voting details from state governments and The Associated Press. However, a study found that 29% of AI responses about voting were incomplete, inaccurate, or outdated. Spanish language responses were even less accurate, often containing confusing translations or missing citations. AI models struggled with specific questions about ballot harvesting and voting machine security.
Better Provider Data Is Key for Health Systems and AI
Most health systems and plans need better provider data to grow their networks and use AI effectively. Currently, only 42% of health systems and 32% of health plans use this data well for market expansion. Verato, an identity intelligence company, helps fix fragmented records so organizations can know exactly who their providers are. This accurate data allows companies to retire old systems, save money, and improve workflows like claims processing. Experts say organizations must first clean up provider identities before adding advanced network insights to support AI tools.
California Nurses Push for AI Safety Laws
Nurses in California celebrated the final passage of two landmark bills, AB 1979 and AB 2575, that set rules for artificial intelligence in healthcare. These bills are now waiting for Governor Gavin Newsom to sign them into law. The laws aim to make AI systems transparent, fair, and explainable while giving patients the right to appeal AI decisions. The California Nurses Association, led by Assemblymember Ash Kalé and Senator Scott Wiener, argues these rules are needed to stop AI from worsening health biases. The group also wants a new statewide task force to oversee how AI is developed and used in the healthcare system.
Akamai Launches AI Assistant for Web Security
Akamai introduced a new AI Assistant for its Web Security Analytics platform to help security teams investigate threats faster. This tool lets analysts use natural language prompts to navigate data, apply filters, and find relevant documentation without clicking through many menus. The assistant can identify security anomalies, suggest fixes like WAF rule tuning, and guide users through remediation steps. It is designed to support human analysts rather than replace them, ensuring that expert judgment remains essential for security decisions. This feature is the first step in a broader plan to add AI assistance across Akamai's entire application protection platform.
New AI Method Improves Recovery of Critical Materials
Researchers developed a new AI method called Decision-Focused Active Learning to improve the recovery of critical materials like neodymium-iron-boron and samarium-cobalt magnets. This approach uses past results to choose the best experiments, finding optimal outcomes in 16 to 24 tests instead of the usual 48. The study analyzed records from Pacific Northwest National Laboratory to show how this method reduces costs and improves efficiency. The team plans a future test to confirm these results at a larger scale using shared standards for measurements and economic inputs.
Caterpillar Grows Revenue as AI Demand Rises
Caterpillar Inc., known as CAT, reported its biggest quarter in history with a 73% increase in revenue from its construction industries segment. While many focus on computer chips, CAT is benefiting from the AI buildout by supplying heavy machinery and autonomous systems. The company has a long history of innovation and a strong business model that continues to drive growth in the AI infrastructure space. CAT is positioning itself as a key player in the industry by expanding its leadership in autonomous technologies and equipment.
Sources
- Security Agencies Flag Chinese AI Distillation, Push for China Car Ban Accelerates - The US-China Business Council
- China rejects US AI model ‘distillation’ allegations, warns of retaliation
- AI-powered attack exploited PaperCut flaws to hack 395 organizations
- Hundreds of AI agents helped PaperCut attacker hit 395+ orgs, and some went off script
- Voters Are Turning to AI for Election Help. The Answers Aren't Always Right
- Nearly All Health Systems and Health Plans Need Better Provider Data. Few Have It, and AI Investment Is at Risk.
- Nurses celebrate final passage of landmark A.I. bills, urge Gov. Newsom to sign
- Introducing AI Assistant for Akamai Web Security Analytics
- Decision-Focused Active Learning for Scale-Aware Critical-Materials Recovery
- CAT Could Be the Best AI Infrastructure Stock You’re Overlooking
Comments
Please log in to post a comment.