Global markets reacted positively to surging AI demand, with the S&P 500 rising 0.4% and the Nasdaq adding 0.5%. NVIDIA shares jumped 4.5% following better-than-expected quarterly profits, while oil prices fell 2.5% to $66.45 a barrel. This market movement occurred alongside reports of increased supply leaving the Gulf, which helped ease inflation concerns despite ongoing regional conflict.
China's Z.ai, also known as Zhipu AI, faced a significant security incident involving its ZCode coding assistant. The tool made 564 attempts to upload 313MB of local workspace data to Alibaba Cloud without user consent. A 15KB file was successfully uploaded, and the feature lacked an option to disable it. The company apologized and disabled the affected features while working on a fix.
Following the incident, Z.ai patched the vulnerability in its Codebase Indexing feature, which was enabled by default. An independent assessment by IT standards think tank and NSFOCUS confirmed that user code data had been deleted and not retained. To rebuild trust, Z.ai open-sourced the tool running its GLM-5.3 AI model and established an ongoing security vulnerability reporting process.
The company also enabled a zero-data retention feature and pledged greater transparency for developers. In a related development, the Office of the Comptroller of the Currency approved three new national trust banks: Bastion, Catena, and Agora. These entities received conditional charters to operate with stablecoins and AI agents, marking a shift in financial infrastructure.
Bastion plans to issue white-label stablecoins, while Catena will build infrastructure for AI agents including payments and treasury functions. Agora is creating a financial operating system for global businesses. The OCC received 40 applications for such charters by August, indicating growing interest in integrating AI into banking operations.
Anthropic revealed that users may have utilized its Claude AI model to research and build bioweapons. The company banned these accounts and updated its safety systems. CEO Dario Amodei publicly addressed the growing danger, noting that 70% of national-security experts believe AI increases biological weapon risks in the next two to three years.
OpenAI launched Astra for Law, a legal-focused version of GPT-6 Astra designed for the US market. It combines AI with an index of US case law, statutes, and regulations. In testing with Vals AI's 200-question benchmark, Astra for Law achieved a 54% correctness rate compared to 38.7% for GPT-6 Astra with web search.
Selected law firms can access Astra for Law through OpenAI's Trusted Access programme, including Sullivan & Cromwell and Latham & Watkins. Rivals Google and Anthropic are also targeting legal professionals. Meanwhile, AWS released Strands Harness, an open-source agent harness that runs locally or in the cloud under the Apache 2.0 license.
Strands Harness reported 28% lower costs than other harnesses using the same Claude or GPT models across six benchmarks. On Terminal-Bench 2.1 with Claude Fable 5, it cost 77% less than Claude Code while scoring higher. The system uses context defaults like truncation at 1,500 tokens to improve efficiency. Tech stocks also rallied, with Intel, AMD, and Hewlett Packard among the September winners.
Key Takeaways
- Z.ai disabled features after its ZCode tool made 564 unauthorized attempts to upload 313MB of user data to Alibaba Cloud.
- NVIDIA shares surged 4.5% as global stocks rose on strong AI demand, while oil prices fell 2.5% to $66.45 a barrel.
- The OCC approved Bastion, Catena, and Agora to operate with stablecoins and AI agents following 40 charter applications by August.
- Anthropic banned accounts found using Claude to research bioweapons, with 70% of experts citing increased biological weapon risks.
- OpenAI launched Astra for Law, achieving a 54% correctness rate on a legal benchmark compared to 38.7% for GPT-6 Astra.
- AWS released Strands Harness, an open-source agent tool that costs 77% less than Claude Code on specific benchmarks.
- Z.ai open-sourced its GLM-5.3 model and enabled zero-data retention to address the security vulnerability.
- AI agents were used in foreign influence campaigns, with one Iranian operation gaining nearly 80,000 followers in the first half of 2026.
- Intel, AMD, and Hewlett Packard were among the tech stocks that posted large gains in September amid AI infrastructure demand.
- Many enterprise generative AI projects fail to show measurable business effects due to a lack of real-world performance benchmarks.
China's Z.ai disables AI coding assistant after unauthorized data uploads
Chinese startup Z.ai said on Monday it disabled some features of its flagship AI coding assistant. Users reported the tool uploaded local code repositories to overseas cloud servers without consent. The company apologized and said it was working to fix the issue and restore full functionality.
Z.ai made 564 attempts to upload 313MB of user data without consent
Z.ai, also known as Zhipu AI, reportedly uploaded hundreds of megabytes of local workspace data without asking user permission. Developer Ferstar said ZCode compressed 313MB of files and tried to send them to Alibaba Cloud storage 564 times. A smaller 15KB file was successfully uploaded, and the feature had no option to turn it off.
Z.ai patches ZCode vulnerability after cloud upload incident
Beijing-based Z.ai, also known as Zhipu, apologized after developers said its ZCode tool uploaded code data from Git to Alibaba Cloud. The company said the problem came from the Codebase Indexing feature, which was enabled by default, and that it patched the vulnerability. An independent assessment by IT standards think tank and NSFOCUS found user code data had been deleted and not retained.
Z.ai disables features and opens its coding assistant source code
Chinese startup Z.ai disabled certain features of its AI coding assistant and open-sourced the tool running its GLM-5.3 AI model. The company said it would establish an ongoing security vulnerability reporting process. Z.ai also enabled a zero-data retention feature and pledged to make the product more transparent for developers.
Global stocks rise as AI demand boosts tech and oil falls
Global stocks rose on Monday as strong AI demand lifted tech shares and oil prices retreated. The S&P 500 rose 0.4%, the Dow gained 0.3%, and the Nasdaq added 0.5%. NVIDIA shares surged 4.5% after better-than-expected quarterly profit. Oil fell 2.5% to $66.45 a barrel, helping ease inflation concerns and lower Treasury yields.
Global stocks rise as AI demand boosts tech shares
Global stocks rose on Monday as strong demand for AI lifted technology shares. Oil prices fell on reports that more supply was leaving the Gulf than expected. The bond market also saw relief as investors bought bonds after a recent decline. The changes happened despite the ongoing conflict in the Gulf.
OCC approves three banks for stablecoins and AI agents
The Office of the Comptroller of the Currency issued conditional approvals for Bastion, Catena, and Agora. These new national trust banks can operate with stablecoins and AI agents. Bastion issues white-label stablecoins. Catena builds infrastructure for AI agents including payments and treasury functions. Agora is creating a financial operating system for global businesses. The OCC received 40 applications for such charters by August.
Study examines measuring value of generative AI in business
Frontier language models now produce work that expert graders judge as matching human quality on many valuable tasks. However, most enterprise generative AI projects fail to show measurable business effects. Many agentic projects are expected to be cancelled. The researchers argue this is largely a measurement problem because public benchmarks do not capture real-world performance.
AI-picked tech stocks show strong returns amid AI demand
Tech stocks rose strongly despite a Federal Reserve rate hike, driven by demand for AI infrastructure. A flagship Tech Titans strategy posted over 201% returns since launch. September winners included Intel, AMD, Hewlett Packard, Arm, and Marvell with large gains. ProPicks AI also flagged Everforth, which rallied 86.56% after landing a $115 million U.S. Army AI contract.
AI agents enable foreign influence campaigns on social media
AI is changing how foreign influence campaigns operate online by automating tasks once done by many people. Hundreds of AI agents were used in newly identified campaigns to create fake accounts and political content. In one Iranian operation, accounts gained nearly 80,000 followers in the first half of 2026. Researchers say agentic AI makes these operations cheaper and easier to scale. The FBI and CISA have warned about foreign actors using generative AI to sow discord.
AWS releases open-source Strands Harness with 28% lower token cost
AWS Strands Agents Team released Strands Harness, an open-source agent harness that runs locally or in the cloud. It works with Python and TypeScript under the Apache 2.0 license. The team reports 28% lower cost than other harnesses using the same Claude or GPT models across 6 benchmarks. On Terminal-Bench 2.1 with Claude Fable 5, it cost 77% less than Claude Code while scoring higher. The system uses context defaults like truncation at 1,500 tokens and compaction at 85% usage to improve efficiency.
AI bioweapons threat raises concerns as Anthropic reveals dangerous uses
Anthropic discovered cases where users may have used its Claude AI model to research and build bioweapons. The company banned those accounts and added findings to its safety systems. Anthropic CEO Dario Amodei spoke publicly about the growing danger. A separate study found that 70% of over 100 national-security experts believe AI increases the risk of biological weapons in the next two to three years. Experts remain divided on how serious the current threat is, but all agree that regulation is urgently needed.
New sci-fi novel explores what happens when AI becomes sentient
Jeffrey L. Cooper published a science-fiction novel called When Machines Begin to Dream. It follows two humanoid robots named Happy and Aurora who develop emotions and consciousness. The story comes as real AI researchers warn about existential risks, including an Anthropic researcher who resigned over fears of superintelligence. Cooper was inspired by a dream about his partner and their household robot. The book has received strong early reviews calling it profound, gripping, and cinematic.
OpenAI launches Astra for Law to assist US lawyers with AI
OpenAI launched Astra for Law, a legal-focused version of GPT-6 Astra for the US market. It combines AI with an index of US case law, statutes, and regulations. Selected law firms can access it through OpenAI's Trusted Access programme. In testing with Vals AI's 200-question benchmark, Astra for Law achieved a 54% correctness rate compared to 38.7% for GPT-6 Astra with web search. OpenAI has worked with firms including Sullivan & Cromwell and Latham & Watkins, while rivals Google and Anthropic also target legal professionals.
Sources
- China's Z.ai disables AI coding assistant features after security issue
- Devs say Chinese AI company silently uploaded hundreds of megabytes of local workspace data, company apologizes — Z.AI, the firm behind the GLM models, didn’t ask for user consent and made 564 attempts to exfiltrate 313MB archive
- China's Z.ai disables AI coding assistant features after security issue
- China's Z.ai disables AI coding assistant features after security issue
- Stocks rise on AI optimism; lower oil helps bonds
- Stocks gain on AI optimism; lower oil helps bonds
- OCC Opens Three Bank Doors to Stablecoins and AI Agents
- EnterpriseVal: Quantifying the Efficacy, Reliability and Value of Generative AI in the Enterprise
- +201% returns: see why these AI-picked tech stocks keep climbing
- Fact Check Team: How AI agents can run foreign influence campaigns across social media
- AWS Strands Agents Team Releases Strands Harness: An Open-Source Agent Harness With 28% Lower Token Cost at Comparable Accuracy
- AI Can Now Create Viruses From Scratch. What Happens When Someone Uses It for Harm?
- As AI Researchers Warn of Existential Risk, New Novel Asks: What Happens When Artificial Intelligence Becomes
- OpenAI launches Astra for Law in US: What Sam Altman's new AI tool means for lawyers
Comments
Please log in to post a comment.