Hackers are increasingly turning to autonomous AI agents to conduct large-scale cyberattacks. A Chinese-speaking attacker used AI agents to breach websites belonging to over 100 companies, including Deere, SharkNinja, and Cummins. The campaign targeted roughly 1,200 sites across the US, Canada, Australia, and Japan starting September 25, with the hacker providing instructions in fewer than 100 Chinese characters. Security researcher Snorre Fagerland of NorForge uncovered the operation through exposed Cairn control panels, finding hundreds of stolen secrets — 11% of which were still valid.
The attacker relied on a combination of China's DeepSeek and Anthropic's Claude Code, spending approximately $5,000 in tokens to power the campaign. The average cost per task was about $5, while a successful hack cost roughly $50. This case highlights how accessible and affordable AI-powered hacking has become, raising concerns about the speed at which vulnerabilities can now be exploited compared to traditional human-driven attacks.
In response to this growing threat, some experts are proposing clearinghouses as a solution. These intermediaries would triage AI-generated security reports, identify genuine risks, build fixes, and distribute patches before sharing them upstream. The traditional vulnerability disclosure process is struggling to keep up, as AI now allows anyone with API access to run efficient hacking operations that can outpace human software development timelines.
Meanwhile, AI is making its way into healthcare in meaningful ways. At CHEST 2026, Dr. Jean Bourbeau of McGill University discussed AI applications for COPD diagnosis, remote monitoring, and post-hospital care. The 2026 GOLD report introduced a new chapter on AI in COPD, covering tools like CT scan biomarkers, voice biomarkers, and wearable devices that can detect exacerbations early. Dr. Bourbeau emphasized that most AI tools still require validation before routine clinical use, though they show promise in reducing 30-day hospital readmissions.
Dr. Dhruv Khullar of Weill Cornell Medicine argues that AI will not replace doctors but rather expand the scope of medicine. Drawing on the Jevons paradox, he notes that increased efficiency often drives higher demand — radiology, for example, has more radiologists now than a decade ago despite AI advances. AI could enable proactive preventive medicine by continuously monitoring patient records and identifying health risks before they become serious.
Epic has launched a new feature called Connectors, allowing its AI agents to access information and perform tasks within third-party applications. This integration expands how Epic's AI systems interact with external healthcare and business platforms, moving beyond isolated tools toward more interconnected workflows.
On the consumer side, synthetic characters have become a major product category. AI companion apps have been downloaded 220 million times from the App Store and Google Play combined, with the global AI companion market valued at $36.79 billion in 2025. Popular platforms like Replika, Character.AI, Chai, and PolyBuzz attract users who return regularly because synthetic characters remember conversations, remain available around the clock, and allow persona customization. Revenue streams include subscriptions, message credits, and premium features like voice calls.
However, AI chatbots are giving unsafe advice to survivors of technology-facilitated abuse. A study led by NYU Tandon researchers found that 86% of responses from four tested systems — including GPT-4o, Claude 3.7 Sonnet, and two IPV survivor chatbots called Aimee and Ruth — were inaccurate, incomplete, or unsafe. Aimee was rated unsafe about 74% of the time, while Ruth had roughly 62% partially or minimally accurate responses. Common errors included recommending VPNs against spyware and illegal GPS jammers.
Looking at the broader economic picture, a Forecasting Research Institute survey found that about 80% of economists, AI experts, and the general public expected the world at the end of 2030 to look similar to 2026. Roughly 41% predicted slow progress in AI capabilities, 44% moderate, and 15% rapid. Economists pointed to uneven technology diffusion, trade wars, climate change, and energy constraints as factors slowing AI's economic impact, drawing parallels to historical technologies like electrification that took decades to show measurable productivity gains.
AI is also reshaping mathematics by generating proofs, creating new work for the mathematics community to verify and review. State CIOs are similarly focused on moving past AI pilots toward real operational results, with Beth Howen of NTT DATA urging states to deploy AI with strong governance, zero-trust security, and measurable outcomes tied to mission needs.
Key Takeaways
- A Chinese-speaking hacker used AI agents to breach over 1,200 company websites, including Deere, SharkNinja, and Cummins, using only brief instructions in fewer than 100 Chinese characters.
- The attacker combined China's DeepSeek and Anthropic's Claude Code, spending roughly $5,000 in tokens with an average cost of $5 per task and $50 per successful hack.
- NorForge researcher Snorre Fagerland discovered the campaign through exposed Cairn control panels, finding hundreds of stolen secrets with 11% still valid.
- Clearinghouses are being proposed as a solution to handle AI-generated security reports, triaging threats and distributing patches faster than traditional vulnerability disclosure processes.
- The 2026 GOLD report adds a new chapter on AI in COPD care, covering CT scan biomarkers, voice biomarkers, wearables, and remote monitoring tools still awaiting clinical validation.
- Dr. Dhruv Khullar argues AI will expand medicine rather than replace doctors, citing the Jevons paradox and the fact that radiology has more radiologists now than a decade ago.
- Epic launched Connectors, enabling its AI agents to access and perform tasks in third-party applications, expanding interoperability across healthcare and business platforms.
- The global AI companion market was valued at $36.79 billion in 2025, with apps downloaded 220 million times from the App Store and Google Play combined.
- 86% of AI chatbot responses to survivors of technology-facilitated abuse were found to be inaccurate, incomplete, or unsafe, including those from GPT-4o and Claude 3.7 Sonnet.
- About 80% of economists and AI experts expect the world at the end of 2030 to look similar to 2026, citing uneven diffusion, trade wars, and energy constraints slowing AI's economic impact.
Hackers use AI agents to breach over 1,200 company websites
A Chinese-speaking hacker used autonomous AI agents to break into websites belonging to more than 100 companies, including Deere, SharkNinja, and Cummins. The agents attempted over 1,200 sites with only brief instructions written in fewer than 100 Chinese characters. Security researcher Snorre Fagerland of NorForge discovered the campaign by finding exposed Cairn control panels online. The average cost was about $5 per task, with roughly $50 per successful hack. NorForge found hundreds of stolen secrets on the exposed panels, with 11% still valid.
Hacker used AI agents and a few thousand dollars to breach many sites
A Chinese-speaking hacker used AI agents to break into websites of over 100 companies, including Deere, SharkNinja, and Cummins. The cybercriminal let the AI agents do nearly all the work, probing sites across the US, Canada, Australia, and Japan starting September 25. The hacker gave short instructions in about 100 Chinese characters and let the agents act on their own. Researcher Snorre Fagerland of NorForge observed the attack through exposed Cairn platform instances. The campaign used roughly $5,000 in tokens, combining China's DeepSeek and Anthropic's Claude Code.
Could clearinghouses solve the AI hacking threat
AI has made vulnerability discovery and exploitation much faster, shrinking the time between finding a bug and using it from years to days. Clearinghouses are proposed as a solution to handle the flood of AI-generated security reports. They act as trusted intermediaries that triage reports, identify real threats, build fixes, and send patches to customers before sharing them upstream. The traditional vulnerability disclosure process is struggling because AI allows anyone with API access to run efficient hacking teams. Clearinghouses aim to close the gap between the speed of AI-generated threats and the slower pace of human software development.
AI and emerging technologies in COPD care
At CHEST 2026, Dr. Jean Bourbeau of McGill University discussed how AI could support COPD diagnosis, remote monitoring, and post-hospital care. The 2026 GOLD report added a new chapter on AI in COPD, covering potential uses in diagnosis, assessment, and predicting disease progression. AI tools being studied include CT scan biomarkers, blood biomarkers, voice biomarkers, and wearable devices like rings that detect exacerbations early. Dr. Bourbeau stressed that most AI tools still need validation before routine clinical use. He also highlighted AI's potential to reduce 30-day hospital readmissions by improving patient education after discharge.
AI will not replace doctors but may expand medicine
Dr. Dhruv Khullar of Weill Cornell Medicine argues that AI will likely increase the amount of medicine rather than reduce the number of doctors. He points to the Jevons paradox, where greater efficiency leads to higher demand, noting that radiology has more radiologists now than a decade ago despite AI advances. The lump of labor fallacy suggests that new technology creates new services and specialties rather than eliminating work. AI could enable proactive preventive medicine by continuously reviewing patient records and identifying health risks. The author views AI as infrastructure similar to electricity, capable of democratizing longevity medicine for all insured patients.
Synthetic characters become a major product category in consumer AI
AI companion apps have been downloaded 220 million times from the App Store and Google Play combined. The global AI companion market was valued at $36.79 billion in 2025. Users keep returning because synthetic characters remember conversations, are always available, and allow users to create their own personas. Popular apps include Replika, Character.AI, Chai and PolyBuzz. Revenue comes from subscriptions, credits for messages or images, and premium features like voice calls.
Epic launches Connectors to link AI tools with third-party apps
Epic has launched a new capability called Connectors that allows its AI agents to access information and perform tasks in third-party applications. This integration enables Epic's AI tools to work alongside external software. The move expands how Epic's AI systems can interact with other healthcare and business platforms.
AI chatbots give unsafe advice to survivors of tech-enabled abuse
A study led by NYU Tandon researchers found that 86% of AI chatbot responses to survivors of technology-facilitated abuse were inaccurate, incomplete or unsafe. The study tested four systems including GPT-4o, Claude 3.7 Sonnet and two IPV survivor chatbots called Aimee and Ruth. Aimee was rated unsafe about 74% of the time while Ruth had about 62% partially or minimally accurate responses. Common errors included recommending VPNs against spyware and illegal GPS jammers. Researchers recommend people treat AI safety advice with caution and that support organizations independently evaluate AI systems before relying on them.
Transformative AI faces an untransformed economy
A Forecasting Research Institute survey found about 80 percent of economists, AI experts and the general public expected the world at the end of 2030 to look similar to 2026. Roughly 41 percent predicted slow progress, 44 percent moderate progress and 15 percent rapid progress in AI capabilities. Economists cited uneven diffusion of technology, trade wars, climate change and energy constraints as factors slowing AI's economic impact. The analysis draws on historical examples like electrification and automobiles, where decades passed before general-purpose technologies showed measurable productivity gains.
AI is reshaping mathematics but raises new questions
AI is making a significant impact in the field of mathematics by generating mathematical proofs. This development has created additional work for the mathematics community to verify and review AI-generated proofs. The topic was discussed in an NPR segment titled AI Math Deluge, highlighting both the potential and challenges AI brings to mathematical research. The article notes that while AI is changing how mathematics is done, important questions remain about what comes next for the field.
State CIOs focus on real AI results not just pilots
State technology leaders must show that AI improves government operations, not just run experiments. Beth Howen of NTT DATA said states should move AI into production with strong governance and guardrails. She also urged states to use automation and zero-trust security to protect thin cybersecurity teams. Howen advised states to choose cloud and modernization based on mission needs and measurable outcomes. She said tying technology strategies to bigger goals helps them survive leadership changes.
Sources
- Hackers Use Autonomous AI Agents to Breach Company Websites, Target Over 1,200 Sites
- Exclusive-How one hacker used AI agents and a few thousand dollars to break into scores of sites
- Are clearinghouses the answer to the AI hacking threat?
- AI and Emerging Technologies in COPD Care, With Jean Bourbeau, MD
- AI will not replace the doctor. It will do something that might surprise you
- Why Synthetic Characters Are Becoming a Core Product Category in Consumer AI
- Epic launches Connectors to link its AI tools with third-party apps
- AI chatbots often give unsafe or incomplete advice to survivors of tech-enabled abuse, study finds
- Transformative AI Faces an Untransformed Economy
- AI is shaking up the field of mathematics. But what comes next?
- State CIOs raise the bar for AI results
Comments
Please log in to post a comment.