AI agents are spreading across organizations, but their growing reach raises questions about oversight and human expertise. Nudge Security reports that more than half of organizations using agents run them on at least two platforms, including Claude and OpenAI; half of those organizations run seven or more agents, and over four in ten run more than 10. Anthropic integrations appear at 74% of organizations in the report and OpenAI integrations at 67%. Eighteen percent of observed prompts contain pasted content or uploaded files. Nudge recommends tracking agents and their access, reviewing OAuth connections and browser extensions, and checking what data tools can reach.
Several articles warn that automation could weaken the experience people need to make sound decisions. In cybersecurity, AI can analyze alerts and summarize incidents, but junior analysts still need practice with phishing and suspicious activity; one proposed approach is to have them form an independent view before consulting AI. A separate article raises a similar concern for Gen Z: AI may speed up work while reducing entry-level opportunities to learn how professions operate and how to recognize faulty AI output. OpenAI research chief Mark Chen and Thinking Machines chief scientist John Schulman also say agents can run research tests, sometimes for days, but people must assess results and redirect work when experiments miss the mark. Chen says models still lack research taste. Thinking Machines is also working on low-precision training in End-to-End 4-Bit Training.
AI’s risks extend to children, public infrastructure and investors. In a Nov. 17, 2026 commentary, Soo Battle warns that realistic deepfakes can enable bullying and harassment, while chatbot interactions with minors may include sexual material or encouragement of dangerous actions. Another article urges Congress and state regulators to protect households and small businesses from paying for electricity upgrades serving AI data centers. Long-term contracts and upfront payments could help allocate costs, although the Federal Energy Regulatory Commission cannot require states to act under current law; questions remain about facilities with their own power but grid backup needs. Morningstar’s 2026 survey found ESG considerations applied to 44% of global assets, unchanged year over year, while concern about AI’s environmental impact rose from 12% to 25% of respondents. Nearly six in 10 worried AI demand could increase energy costs and inflation.
Other reports describe efforts to make AI systems safer and more useful. Japan and ASEAN agreed on Nov. 17, 2026, to launch an AI safety and security scheme, though details were not provided. FinVerify checks financial agents’ numerical claims against evidence using fixed checks for reporting periods, sources, units and accounting rules. Across five benchmarks, it achieved 76.7% accepted-claim precision on Agent Finance Reasoning, versus 56.4% for an AI judge, while covering 81.2% of claims; results varied where authoritative evidence was limited. In biotechnology, AI suggested DNA changes for a bacteriophage designed to attack harmful bacteria. Radical Numerics says it used an AI-designed virus in a patient, but the treatment has not undergone clinical trials or received full FDA approval, and the outcome and administration details remain undisclosed.
Australia is also being urged to build AI expertise beyond products developed by large technology firms. The article backs proposed security rules for frontier AI companies while calling for local development and testing, predictable regulation and infrastructure investment from companies including Microsoft, Amazon, Anthropic and OpenAI. It cites researchers Yoshua Bengio and Geoffrey Hinton as examples of AI’s research roots, and argues that a possible investment bubble would not make the technology itself a bubble. Separately, the sustainable-investing report says investors are increasingly assessing AI-related energy use, electricity costs, dependence on a small group of companies and infrastructure providers, ethics, automation’s impact on jobs and workforce transitions.
Key Takeaways
- Nudge Security found that more than half of organizations using AI agents operate them on at least two platforms; 74% have an Anthropic integration and 67% an OpenAI integration.
- AI can accelerate cybersecurity and research work, but junior analysts and researchers still need hands-on practice and human judgment to assess outputs and redirect work.
- AI may reduce entry-level learning opportunities for Gen Z, while young workers’ use of AI tools can also raise security concerns.
- Soo Battle warned on Nov. 17, 2026, that deepfakes and chatbot interactions can expose children and teens to harassment or harmful content.
- Japan and ASEAN agreed on Nov. 17, 2026, to launch an AI safety and security scheme; the report offered no further details.
- FinVerify reached 76.7% accepted-claim precision on Agent Finance Reasoning versus 56.4% for an AI judge, with 81.2% claim coverage.
- AI-assisted bacteriophage design has been used in a patient by Radical Numerics, but the treatment lacks clinical-trial testing and full FDA approval; its outcome is unknown.
- Congress and state regulators are urged to ensure data centers pay for electricity infrastructure serving them; FERC cannot mandate state action under current law.
- Morningstar found concern about AI’s environmental impact rose from 12% to 25% year over year, and nearly six in 10 investors worried AI demand could raise energy costs and inflation.
- Australia is urged to develop and test AI locally, attract investment from Microsoft, Amazon, Anthropic and OpenAI, and draw on research traditions associated with Yoshua Bengio and Geoffrey Hinton.
Cybersecurity AI Must Train Analysts, Not Replace Their Judgment
AI can speed up cybersecurity investigations by analyzing signals, summarizing incidents, and handling repetitive tasks. But junior analysts still need hands on experience with phishing emails, alerts, and suspicious activity to build pattern recognition and judgment. The article recommends having analysts form their own view before comparing it with AI, so the technology can coach rather than give answers too early. Security leaders should decide which tasks AI can take over and which experiences employees still need to become skilled decision makers.
AI Could Remove the Early Jobs Gen Z Needs to Learn at Work
AI can help young professionals write, develop ideas, and finish tasks faster, but it may also remove entry level jobs where people learn their professions. The article argues that junior workers need hands on experience to understand how work is done and to spot when an AI answer is wrong. Gen Z’s experience includes digital childhoods, pandemic disruptions to school and work, and worries about job security after financial crises and layoffs. The article warns that losing early career learning opportunities could weaken trust in education and employers, while young workers’ use of AI tools may also create security concerns.
Nudge Security Finds AI Agents Spread Across Multiple Platforms
A Nudge Security report found that more than half of organizations using AI agents run them on at least two platforms, including Atlassian, Claude, Cursor, OpenAI, and Zapier. Among these organizations, half run seven or more agents, and more than four in ten run over 10. The report also found that 74% have at least one Anthropic integration and 67% have an OpenAI integration, while 18% of observed prompts include pasted content or an uploaded file. The findings use anonymized, aggregated customer data, with adoption and integration information from September 2026 and agent data from October 2026. Nudge Security recommends tracking agents and their access, reviewing OAuth connections and browser extensions, and focusing on what AI tools can reach.
Japan and ASEAN Agree to Launch AI Safety and Security Scheme
Japan and ASEAN agreed on November 17, 2026, to launch a scheme focused on AI safety and security. The report provides no further details about the scheme.
AI Helps Design Virus to Fight Harmful Bacteria
AI helped scientists design a bacteriophage, a virus that attacks harmful bacteria, by suggesting changes to the virus’s DNA. Scientists can adjust the phage’s “legs” so it can attach to a specific bacterial strain, and AI can suggest possible genetic changes for lab testing. Northeastern biotechnology professor Sudhakaran Prabakaran explained how the approach works, while biotech company Radical Numerics reported using an AI designed virus in a patient. The treatment has not gone through clinical trials or received full FDA approval, and its reported outcome is unknown. The company has not shared details about how the virus was given to the patient, citing privacy concerns.
FinVerify checks financial AI claims against evidence
FinVerify is a framework that checks numerical claims made by financial agents against supporting evidence. It uses an AI model to organize claims and evidence, then applies fixed checks for issues such as reporting periods, sources, units and accounting rules. In tests across five benchmarks, it raised accepted-claim precision to 76.7% on Agent Finance Reasoning, compared with 56.4% for an AI judge, while covering 81.2% of claims. It reached 99.8% precision on XBRLFiling and 87.9% on FinanceBench-67, but gains were smaller on FinQA and TAT-QA, where authoritative evidence was limited.
Soo Battle warns that AI can expose children to harm
In a Nov. 17, 2026 commentary, Soo Battle warns that artificial intelligence creates new risks for children and teens. Battle, who has helped child sexual abuse victims in Central Texas for 11 years, says AI can make realistic deepfakes used for bullying and harassment. She also warns that chatbots may have inappropriate or harmful conversations with minors, including sexual discussions or encouragement of dangerous actions. Battle urges parents and other adults to stay alert, noting that AI safety safeguards remain limited.
New Congress urged to shield households from AI data center costs
The article says the new Congress should respond to the effects of AI data centers, including the risk that residential and small business customers will pay for related electricity upgrades. State regulators can use cost allocation and rate design to make data centers pay for generation, transmission and distribution investments that serve them. Long-term contracts and upfront payments could also help protect customers if a data center uses less power or closes. The article notes that the Federal Energy Regulatory Commission can encourage state action but cannot require it under current law. It also raises unresolved questions about data centers with their own power supplies, since many still rely on the grid for backup and peak demand.
AI agents can test research ideas, but scientists still guide them
OpenAI research chief Mark Chen and Thinking Machines chief scientist John Schulman say AI agents can run research tests, but people still provide research judgment. Schulman described agents running trials for days that produced reasonable results but missed the mark, requiring him to review the findings and redirect the work. Chen says research taste means knowing which ideas will remain useful and connect with other ideas, a skill models still lack. The article also discusses automated benchmark creation, small tests that can predict larger results, and Thinking Machines Lab’s work on low-precision training in End-to-End 4-Bit Training.
Australia should build AI expertise beyond Big Tech
The article argues that Australia should treat AI as a major discovery, not just a product from large technology companies. It says the government’s proposed security rules for frontier AI companies are worthwhile, but Australia must also develop and test AI locally if it wants to shape the technology. That would mean attracting infrastructure investment from companies such as Microsoft, Amazon, Anthropic and OpenAI. The article points to university researchers Yoshua Bengio and Geoffrey Hinton as examples of AI’s roots in research, and says a possible investment bubble would not mean the technology itself is a bubble. It calls for sensible, predictable and consistent regulation while Australia builds a role in AI development.
AI reshapes the risks investors consider in sustainable investing
Morningstar’s 2026 survey found that 44% of assets worldwide included ESG considerations, unchanged from the previous year, while regulation played a smaller role in ESG adoption. In Europe, the share of asset owners applying ESG to more than 75% of assets under management rose from 5.5% in 2022 to almost 20%. Investors are paying more attention to risks from generative AI, including its effects on data centre energy use, electricity costs, inflation, and reliance on a small number of companies and infrastructure providers. Concern about AI’s environmental impact rose from 12% to 25% of respondents year over year, and nearly six in 10 worried that AI-related demand could raise energy costs and inflation. Investors are also examining AI ethics, automation’s effects on jobs, and workforce transitions.
Sources
- The AI Shortcut Cybersecurity Leaders Should Avoid
- AI Is Erasing the Entry-Level Years Gen Z Needs to Build Judgment
- More Than Half of Organizations Using AI Agents Run Them Across Multiple Platforms, Nudge Security Research Finds
- BREAKING NEWS: Japan, ASEAN agree on launching scheme for AI safety, security
- This time, the virus is on the patient's side. AI helped design it
- FinVerify: A Verification Framework for Financial Agents
- Opinion: Stop artificial intelligence from abusing our children
- What the new Congress needs to do now on AI data centers
- AI Research Interns Run Tests, Humans Keep Taste
- To make good decisions on AI, stop reducing it to Big Tech
- AI ushers in a more mature sustainable investment case
Comments
Please log in to post a comment.