AWS Adds Full Fine Tuning to SageMaker AI Serverless

The European Commission says an AI-generated cybercampaign targeted its systems, detected on October 22, and disrupted internal services for more than ten days. Sysper, used for payroll, leave and timesheets, and MyPMO, which handles medical reimbursements, were among the affected platforms; translation and staff health-insurance services were also disrupted. Salaries arrived on time, though some employees could not file timesheets or request leave. The Commission says it has no evidence of a breach affecting staff as a whole. Investigators have not identified the attackers or confirmed data theft. ARTEX, a tool connecting to an external AI model, is suspected in some reports but has not been confirmed as the method. The incident follows January and March security events; the March breach reportedly involved about 91.7 GB of compressed data.

Nvidia’s strong quarterly growth outlook helped steady AI stocks after a sharp downturn. CEO Jensen Huang says the company sees continued growth and that its technology is being adopted in healthcare, finance and transportation. Nvidia shares traded near $600, down from a February high above $1,000, amid broader market weakness, concerns about GPU demand and stronger competition. Huang says Nvidia will continue investing in research and development.

Amazon Web Services announced full fine tuning for SageMaker AI Serverless Model Customization on November 18, 2026. In a coding test with more than 100,000 training examples, AWS reported 90.0% semantic correctness, reached after three epochs. Results varied by settings: aggressive learning rates caused substantial losses in general knowledge, and fine tuning did not beat LoRA or the base model on every measure. Supported open models include DeepSeek R1 Distill Qwen 1.5B, Google Gemma 4 26B A4B and NVIDIA Nemotron 3 Nano 30B (A3B).

Researchers introduced Gandalf, an open-source tool for checking AI agents against natural-language task criteria. Its BankerVerifierBench dataset contains 2,464 human-judged criteria drawn from 24 investment-banking tasks. Gandalf matched or outperformed existing tools on F1 scores at lower cost, including tests with Gemini 3 Flash and GPT-5.4, and posted the highest F1 score in the OpenClaw productivity environment.

Tencent’s revenue rose 8% to RMB207.3 billion in the September 2026 quarter, missing forecasts, while net income barely grew. Founder Pony Ma has called on half of the company’s 100,000 employees to use AI to find growth, with the rest focused on efficiency. Tencent has released its Hy4 model and reports 32 million monthly active users for WorkBuddy in September, but its chatbot trails ByteDance’s Doubao. The company is also developing WeChat’s Xiaowei agent, tested since summer, as it contends with AI competition and higher chip and data-center costs.

SpaceX may offer part of its planned $40 billion debt financing for AI chips as soon as this week, Bloomberg reported. Goldman Sachs and Apollo Global Management are expected to underwrite the deal. An Apollo-backed Atlas SP vehicle would purchase Nvidia chips and issue asset-backed securities, keeping the debt off SpaceX’s balance sheet. The proposed sale would test investor appetite for financing backed by AI hardware.

D-Wallet is developing an AI agent that would turn users’ trading requests into verifiable strategies and execute them through supported decentralized exchanges and aggregators. The system is intended to monitor conditions, carry out transactions and track activity within funds and permissions authorized by users, with pause controls and human intervention. The company has not announced a launch date or availability.

Roze AI, founded in 2015 and focused on fire forecasting and wireless fire detection, plans to ring the Nasdaq Opening Bell on November 25, 2026, following its listing. Its Disaster AI Platform is used at government facilities, military bases, production sites and power plants in South Korea; its AI Fire Forecast Alert Program received South Korea’s GS certification in February 2026. The company says the listing will support international expansion. About 70% of its issued and outstanding shares are subject to lock-up agreements.

One article argues that experiential learning can help curricula keep pace with AI by giving students repeated practice in durable workplace skills. It says such learning may also help address opportunity gaps for first-generation students and others with limited access to internships. Separately, Apple is developing agentic AI capabilities to compete with Anthropic, OpenAI and Meta. Job listings reportedly outline an AI assistant that can act on users’ behalf; the report also says Anthropic stock is approaching resistance levels.

Key Takeaways

  • The European Commission’s AI-generated cybercampaign, detected October 22, disrupted internal systems for more than ten days; investigators have not confirmed data theft or identified an attacker.
  • ARTEX is suspected as a possible attack method, but the Commission has not confirmed that it was used.
  • Nvidia shares stabilized near $600 after falling from above $1,000 in February; CEO Jensen Huang says growth and R&D investment will continue.
  • AWS added full fine tuning to SageMaker AI Serverless Model Customization; its reported coding-test result was 90.0% semantic correctness after three epochs, with outcomes dependent on settings.
  • AWS supports DeepSeek R1 Distill Qwen 1.5B, Google Gemma 4 26B A4B and NVIDIA Nemotron 3 Nano 30B (A3B) for the service.
  • Open-source Gandalf matched or exceeded competing verification tools on F1 scores at lower cost, using benchmarks that include 2,464 criteria across 24 investment-banking tasks.
  • Tencent’s September-quarter revenue grew 8% to RMB207.3 billion but missed forecasts; WorkBuddy reported 32 million monthly active users, while its chatbot trails ByteDance’s Doubao.
  • SpaceX may pursue part of a $40 billion AI-chip debt plan through an Apollo-backed vehicle buying Nvidia chips and issuing asset-backed securities.
  • D-Wallet is developing a permission-limited AI agent for on-chain trading, with user controls and no announced launch date.
  • Apple job listings point to an AI assistant that can act for users; Roze AI plans a Nasdaq bell event on November 25, 2026, after its listing.

European Commission cyberattack disrupts staff systems as ARTEX is suspected

A cyberattack disrupted European Commission services for more than ten days, affecting the Sysper human resources platform, MyPMO medical reimbursements, staff health insurance, and translation services. Staff said salaries arrived on time, but some could not submit timesheets or request leave. The cybersecurity team CERT-EU detected the activity, and ARTEX, a tool that connects to an outside AI model, is suspected but has not been confirmed as the attack method. Investigators have not said whether attackers stole staff data or identified who was responsible. The Commission also faced security incidents in January and March, including a March breach that stole about 91.7 GB of compressed data.

Cyberattack disrupts European Commission services for more than ten days

A cyberattack disrupted European Commission internal systems for more than ten days and may have exposed staff data. The Commission temporarily disconnected services including Sysper, which handles payroll, leave requests, and timesheets, and MyPMO, which manages medical reimbursements. The public AI based tool ARTEX is suspected of being used, but investigators have not confirmed a data theft or identified the attackers. Spokesperson Thomas Regnier said the Commission took steps under its usual protocols, while officials criticized limited internal updates. The incident follows a January intrusion that may have exposed staff names and phone numbers and a March breach linked to the Trivy tool.

European Commission confirms AI based cybercampaign targeted its systems

The European Commission said an AI generated cybercampaign targeted it, marking the first time the bloc has publicly reported an AI based cyberattack. The incident was detected on October 22. Spokesperson Thomas Regnier said there was no evidence of a data breach affecting Commission staff as a whole. He did not name a perpetrator or give more details, but said the Commission had taken the necessary steps in response.

Nvidia growth outlook helps steady AI stocks after sharp drops

Sharp drops in Nvidia and other AI related stocks appeared to ease, helping steady Wall Street after a recent downturn. Nvidia reported strong growth in its latest quarterly earnings, and CEO Jensen Huang said the company still sees strong growth ahead. He said Nvidia technology is being adopted in healthcare, finance, and transportation. The AI industry has faced lower GPU demand and increased competition, while Nvidia continues to invest in research and development.

Nvidia shares steady near $600 as CEO expects continued growth

Nvidia and other AI related stocks appeared to stabilize after a recent market slump. FactSet data showed Nvidia shares trading around $600, down from a high of more than $1,000 in February. CEO Jensen Huang said the company still sees strong growth ahead and will keep investing in research and development. The article says the stock has been affected by the wider market downturn and Nvidia's own financial performance.

AWS adds full fine tuning to serverless SageMaker customization

Amazon Web Services announced full fine tuning for SageMaker AI Serverless Model Customization on November 18, 2026. In a coding test with more than 100,000 training examples, AWS reported 90.0% semantic correctness, with the best score after three training epochs. Tests found that results depended on settings: aggressive learning rates led to major losses in general knowledge, and fine tuning did not outperform LoRA or the base model on every measure. AWS says the service supports open models including DeepSeek R1 Distill Qwen 1.5B, Google Gemma 4 26B A4B, and NVIDIA Nemotron 3 Nano 30B (A3B).

New Gandalf tool checks AI agents against task rubrics

Researchers studied ways to verify AI agents working on open-ended tasks that are judged against natural-language criteria. They created BankerVerifierBench, a dataset with 2,464 human-judged criteria from 24 investment-banking tasks, and built Gandalf, an open-source verifier. Gandalf follows three design principles: reactive verification, alignment with the environment, and domain guidance. It matched or beat existing tools on F1 scores while costing less, including in tests with Gemini 3 Flash and GPT-5.4 models. It also achieved the highest F1 score when tested in OpenClaw, a different personal-productivity environment.

Tencent sales miss forecasts as Pony Ma pushes AI overhaul

Tencent revenue rose 8% to RMB207.3 billion in the September 2026 quarter, below analyst expectations, while net income barely grew. Founder Pony Ma has urged half of the company’s 100,000 employees to use AI to find new growth, while others focus on improving efficiency. Tencent has released its Hy4 model and says its WorkBuddy office tool had 32 million monthly active users in September, but its chatbot still trails ByteDance’s Doubao. Tencent is also counting on WeChat and its Xiaowei agent, which has been in testing since summer, as it faces stronger AI competition and rising costs for chips and data centers.

SpaceX chip financing plan puts AI hardware debt to the test

A portion of SpaceX’s planned $40 billion debt financing for artificial intelligence chips could be offered as soon as this week, Bloomberg reported. Goldman Sachs and Apollo Global Management are expected to underwrite the deal. An Apollo-backed Atlas SP vehicle would buy Nvidia chips and issue asset-backed securities, keeping the debt off SpaceX’s balance sheet and helping preserve its investment-grade credit ratings. The sale would test investor demand for debt backed by AI hardware.

D-Wallet builds AI agent for strategy-based on-chain trading

D-Wallet is developing an AI agent that would turn users’ trading needs into verifiable strategies and carry them out through supported decentralized exchanges and aggregation services. The planned system would monitor conditions, execute transactions, and track activity. It is designed to operate only within the funds and permissions users approve, with controls to pause it and allow human intervention. Development is proceeding in stages, but the team has not provided a launch date or availability details.

Roze AI to ring Nasdaq bell after listing

Roze AI will ring the Nasdaq Opening Bell on November 25, 2026, at Nasdaq MarketSite in Times Square to celebrate its listing. Founded in 2015, the company develops AI fire forecasting and wireless fire detection systems. Its Disaster AI Platform is used at government facilities, military bases, production sites and power plants in South Korea, and its AI Fire Forecast Alert Program received South Korea’s GS certification in February 2026. Roze AI says the listing will support international expansion and new business opportunities. About 70% of its issued and outstanding shares are covered by lock-up agreements that restrict transfers during set periods.

Experiential learning can help make curricula AI proof

The article argues that AI is increasing the value of durable human skills as technical expertise changes. Experiential learning gives students repeated practice with skills that reflect challenges in today’s workplace. It may also help close opportunity gaps for first-generation students and others who have limited access to internships.

Apple plans AI agent as Anthropic stock nears resistance

Apple is developing agentic AI capabilities to compete with Anthropic, OpenAI and Meta Platforms. The report says Apple shared a brief overview of its plans in several job listings posted that week. One listing seeks a senior engineering manager for an AI assistant that can take actions on a user’s behalf. The article also says Anthropic stock is approaching key resistance levels.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

AI cybersecurity European Commission cyberattack AI-generated cybercampaign Nvidia GPU demand AWS SageMaker AI model fine-tuning DeepSeek Google Gemma NVIDIA Nemotron AI agent evaluation Gandalf Tencent AI WorkBuddy AI chip financing SpaceX AI trading agent decentralized finance AI fire forecasting agentic AI

Comments

Loading...