The FBI says hackers have used ARTEX, an open-source framework that connects large language models with security tools, in intrusions targeting finance, government, defense, healthcare and technology organizations since at least August 2026. The system can automate reconnaissance, exploitation attempts, credential theft and data extraction. Microsoft observed attacks linked to Storm-0940 reaching privileged access in under three hours. ARTEX has been associated with information stolen from about 67,000 people at South Korean financial institutions and access to more than 2,250 ConnectWise Cloud Backup customer tenants. Its developer removed the public GitHub repository on October 8 after reports of misuse, but copies remain online. The FBI has not linked ARTEX to reported outages affecting Microsoft 365, the FAA, banks, Gmail, Apple Pay or Marriott. Researchers stress that AI is not acting with independent criminal intent, and advise organizations to review permissions and credentials and correlate activity across identity, network, endpoint and cloud systems.
That risk has sharpened attention on agent permissions as well as prompts. In one study’s vulnerable test subset, instructions telling a model to ignore directions found in content reduced average attack success from 58.2% to 14.3%; off-the-shelf guard models left it at about 44%. The authors caution that the results do not represent the full benchmark and that most tests did not include adaptive attackers. A separate discussion led by StarkWare CEO Eli Ben-Sasson warns that AI can identify security flaws but may produce patches that introduce new vulnerabilities. Ben-Sasson has urged careful review of fixes and preparation for AI-driven attacks, including resilient cryptography such as hash-based systems and ZK-STARKs, which he says matter for blockchains, DeFi and identity systems.
OpenAI is adding Le Figaro’s current reporting and archive to ChatGPT, while the French newspaper retains editorial control. Le Figaro says ChatGPT has 27.4 million monthly unique visitors in France and expects its content to make the assistant more complete and reliable in 2027. In another discussion about AI’s effects, Walter Isaacson and Kent Walker consider what Leonardo da Vinci, Benjamin Franklin and Ada Lovelace can teach about creativity. Isaacson highlights teamwork in major breakthroughs and the continuing importance of curiosity and human judgment.
AI adoption is also shaping business and public policy. LinkedIn reports that members adding “founder” to profiles rose 125% in four years, including 275% among Gen Z; nearly half of Americans and roughly two-thirds of millennials and Gen Z surveyed want to start businesses. Nine in 10 small-business owners surveyed use AI, with some saying digital tools help them serve more clients or start with less capital. The report also finds founders increasingly build careers with multiple income sources, often motivated by freedom, flexibility and personal interests. Separately, an article attributes the US AI advantage to government-industry cooperation, citing DARPA partnerships and National Science Foundation support for research, talent and infrastructure.
Local authorities and legal professionals are weighing safeguards. Portland’s City Council is considering Commissioner Dan Ryan’s proposal to pause new AI contracts until stronger privacy protections are ready; existing contracts would remain unaffected. The city is developing an AI-use framework, and a vote is expected in the coming weeks. A legal-sector panel will examine how AI used in areas such as ediscovery, research, hiring and criminal justice risk assessment can encode or spread human bias, and how attorneys can reduce it. Meanwhile, a November 19 report says Russian intelligence services use generative AI in campaigns targeting European divisions, including fake media-style videos and, in Germany, calls to split the country again into the Federal Republic and the German Democratic Republic.
Key Takeaways
- The FBI says ARTEX has enabled automated intrusions since at least August 2026, with Microsoft observing Storm-0940-linked attacks reach privileged access in under three hours.
- ARTEX has been linked to stolen information affecting about 67,000 people at South Korean financial institutions and access to more than 2,250 ConnectWise Cloud Backup customer tenants; the FBI has not tied it to the cited service outages.
- Security guidance emphasizes limiting AI-agent permissions and reviewing credentials, while prompt defenses reduced attack success in one vulnerable test subset but did not eliminate it.
- StarkWare CEO Eli Ben-Sasson warns AI-generated security fixes can create new vulnerabilities and advocates preparation with resilient cryptography, including hash-based systems and ZK-STARKs.
- OpenAI is bringing Le Figaro reporting and archives into ChatGPT; Le Figaro retains editorial control and cites 27.4 million monthly unique visitors for ChatGPT in France.
- LinkedIn reports “founder” profile additions rose 125% in four years and 275% among Gen Z; nine in 10 surveyed small-business owners use AI.
- An article credits US AI strength partly to government-industry cooperation, including DARPA partnerships and National Science Foundation support for research, talent and infrastructure.
- Portland is considering a pause on new AI contracts until privacy safeguards are ready; existing contracts would not be affected.
- Experts are examining bias in AI-assisted legal work, while a November 19 report describes Russian generative-AI disinformation campaigns targeting divisions in Europe.
- Walter Isaacson and Kent Walker discuss creativity in the AI age, emphasizing teamwork, curiosity and human judgment.
FBI warns hackers are using ARTEX to speed up cyberattacks
The FBI says hackers have used ARTEX, an open-source platform powered by large language models, to target organizations in many sectors since August 2026. The system can automate several stages of an intrusion, and some attacks reached privileged access in less than three hours. ARTEX has been linked to stolen information from about 67,000 people at South Korean financial institutions and access to more than 2,250 ConnectWise Cloud Backup customer tenants. The FBI says the tool appeared on GitHub in July 2026, was removed by its developer on October 8, and continues to spread online. The agency has not linked ARTEX to recent outages affecting services such as Microsoft 365, the FAA, banks, Gmail, Apple Pay, and Marriott.
ARTEX raises questions about AI tools and authorized security testing
The FBI says malicious actors have used ARTEX, an open-source framework that connects large language models with security tools, to automate cyber intrusions. Since at least August 2026, attacks have targeted sectors including finance, government, defense, healthcare, and technology, with some reaching privileged access in less than three hours. ARTEX can coordinate tasks such as reconnaissance, attempts to exploit weaknesses, credential theft, and data extraction. Its original developer described it as a tool for authorized testing and removed the public repository in October after reports of misuse, but copies remain available. The article warns that automated attacks can move faster than many organizations’ response processes, while stressing that AI is not acting with its own criminal intent.
ARTEX AI agents can reach privileged access in under three hours
A report citing ExtraHop’s Reveal Labs says threat actors are using ARTEX, an AI-powered penetration-testing framework, to automate network intrusions. Microsoft observed attacks associated with Storm-0940 moving from initial targeting to privileged access in less than three hours. ARTEX links large language models with specialized agents that can run commands, review results, and choose next steps, using local or cloud-based AI models. Researchers tested an English-language version called ScopeWeaver in a controlled setting to identify network behaviors defenders can monitor. The report urges organizations to review application permissions and credentials and to watch for connected attack patterns across identity, network, endpoint, and cloud systems.
Le Figaro content will appear in ChatGPT
OpenAI is adding Le Figaro’s current reporting and archive material to ChatGPT. Le Figaro will keep full editorial control over its content. The French newspaper says ChatGPT has 27.4 million monthly unique visitors in France and is the country’s most-used AI assistant. Le Figaro expects ChatGPT to become more complete and reliable in 2027 because of its content.
LinkedIn finds more Americans are interested in starting businesses
A LinkedIn report found rising interest in entrepreneurship, with the number of members adding “founder” to their profiles up 125% in four years and up 275% among Gen Z. A LinkedIn and Censuswide survey found nearly half of Americans and about two-thirds of millennials and Gen Z want to start a business. Nine out of 10 small business owners surveyed use AI, and some said AI and digital tools helped them serve more clients or start with less capital. LinkedIn research director Sharat Raghavan said the tools can help with tasks such as marketing plans and websites, while business owners still need to hire to grow. The report says founders are increasingly building “portfolio careers” with multiple sources of income, and that freedom, flexibility, and personal interests often motivate them more than potential earnings.
Government and Industry Cooperation Gives the US an AI Edge
The article says the United States’ AI advantage comes from cooperation between government and industry, not just strong research or many startups. DARPA works with private companies, universities, and research institutions on AI for military use, while the National Science Foundation funds research and supports AI talent and infrastructure. The government also backs research centers, education and training, startups, and small businesses. The article says these efforts build a culture of innovation and a pipeline of AI talent.
Eli Ben-Sasson Says AI Can Find Security Bugs but Struggles to Fix Them
StarkWare CEO Eli Ben-Sasson questioned in a November 18, 2026 post on X why AI can find hidden security flaws but cannot reliably repair them. AI-generated patches can introduce new exploitable vulnerabilities, so teams should carefully review proposed fixes before using them. Ben-Sasson also warned on November 5 that AI-driven attacks could emerge before quantum threats, and urged preparation with resilient cryptography, including hash-based systems and ZK-STARKs. He said the risks matter for blockchains, DeFi protocols, and identity systems, while the article notes that StarkWare is built around a technology he recommends.
Portland Weighs Pausing New AI Contracts Until Privacy Rules Are Ready
Portland City Council is considering Commissioner Dan Ryan’s proposal to stop signing new AI contracts until stronger privacy safeguards are in place. The proposal responds to concerns about surveillance and data collection, and would not affect existing contracts. The city’s technology department is developing a framework for AI use, while commissioners say the city should ensure its technology is transparent and accountable. The council is expected to vote in the coming weeks. San Francisco passed a similar resolution in 2020.
Russia Uses AI Disinformation to Deepen Divisions Across Europe
A November 19, 2026 report says Russian intelligence services are using generative AI in information campaigns aimed at dividing societies and destabilizing European Union countries. In Central Europe, fake videos that imitate reputable media outlets spread false claims about politicians, elections, and international events. The campaigns also include fabricated accusations and statements meant to discredit political figures and forces. In Germany, a campaign calls for the country to be divided again into the Federal Republic of Germany and the German Democratic Republic, a move the report says seeks to deepen social divisions and weaken European unity.
AI Agent Security Depends on Permissions as Well as Prompts
The article argues that agent security should focus on what a compromised AI agent is allowed to do, not only on whether a model can be tricked. In one study’s vulnerable test subset, instructions telling the model to ignore directions found in content cut the average attack success rate from 58.2% to 14.3%, while off-the-shelf guard models left it at about 44%. The study’s authors warned that those results should not be applied to the full benchmark, and task success varied by model. The article says prompt defenses are useful but incomplete, since about one in seven attacks still succeeded in the test subset and the study mostly did not test adaptive attackers.
Walter Isaacson explores curiosity and creativity in the AI age
On the latest episode of Dialogues on Technology and Society, Kent Walker talks with biographer and historian Walter Isaacson. They discuss what Leonardo da Vinci, Benjamin Franklin, and Ada Lovelace can teach us about creativity in the age of AI. Isaacson explains why major breakthroughs have often come from teamwork. They also consider why curiosity and human judgment matter when information is close at hand.
Panel examines bias in AI used in legal work
AI tools are used in legal work such as ediscovery, legal research, contract review, hiring, personnel decisions, and criminal justice risk assessment. These tools can encode, hide, and spread human bias rather than remove it. A panel of practitioners and technologists will discuss how bias enters AI assisted legal work and why it can be harder to spot than bias in decisions made only by people. They will also consider practical ways attorneys can reduce bias when using or advising on these tools.
Sources
- What is ARTEX? What to know about FBI's recent AI warning
- ARTEX, AI Penetration Testing and the Rogue Locksmith
- Hackers Use ARTEX AI Agents to Automate Network Intrusions and Gain Privileged Access in Under 3 Hours
- OpenAI puts Le Figaro in ChatGPT
- LinkedIn sees more interest in entrepreneurship, with AI helping founders get their start
- America’s Hidden AI Advantage
- Eli Ben-Sasson questions why AI can find bugs but not fix them
- Portland considers halting new AI contracts
- Artificial intelligence - a tool of hybrid warfare: how Russian intelligence services are trying to divide Europe
- Why AI Agent Security Starts With Permissions, Not Prompts
- Walter Isaacson on curiosity and creativity in the AI age.
- AI and Technology Symposium: Why AI Learns What It Learns, and What
Comments
Please log in to post a comment.