KPMG CEO Tim Walsh Shifts AI Focus to Transformation and Cybersecurity

AI is becoming a source of mental health information and support for young people, but experts caution that access is not the same as care. One article estimates 8.2 million U.S. people ages 12 to 21—nearly one in five—use chatbots for emotional or mental health support. A separate opinion article cites a 2025 Pew survey finding 64% of U.S. teens use chatbots, about three in 10 daily; it also points to a 2026 JAMA Pediatrics survey in which about one in five adolescents and young adults used chatbots for mental health advice, and a Common Sense Media report that nearly three in four teens had used AI companions. Chatbots may help users name feelings, learn about symptoms and prepare to talk with someone, but they can mislead, cannot diagnose or treat conditions, and may reinforce anxiety or obsessive-compulsive symptoms. The authors urge adults to make room for difficult conversations and warn young people against relying on AI for diagnosis, medication decisions or crises, or sharing sensitive information.

At KPMG, CEO Tim Walsh says corporate AI conversations have shifted from fears of mass layoffs toward business transformation and cybersecurity. KPMG’s overall workforce has not declined, though demand is growing for data scientists, data transformation specialists and engineers. Walsh expects some reduction in working hours but says it may be less dramatic than predictions suggest. Because deployment is costly, companies are matching tools to particular jobs rather than giving everyone the same technology. KPMG has launched cybersecurity services, including support for companies in Anthropic’s Cyber Verification Program as they deploy Mythos.

Emergency medical services face a different AI challenge: tools used without agency approval can expose protected health information. EMS attorneys Doug Wolfberg and Ryan Stark say HIPAA still applies to smart glasses, recording devices, ambient listening apps and generative AI. They recommend oversight, updated policies, staff training and safeguards rather than an outright ban. Their discussion also covers recording-consent laws, AI-generated patient care reports, cybersecurity, backup communications and continuity planning, and urges agencies to review policies and test emergency plans.

Governance and security are also prominent in international discussions. The AI and Governance Leadership Forum and Awards 2026, held November 20 at The Ritz-Carlton Abu Dhabi, Grand Canal, brought officials and industry experts together to discuss scaling AI in business and public services. Sessions addressed human oversight, transparency, cybersecurity, agentic AI, third-party models, non-human identities, supply chains and data sovereignty. Separately, UN Secretary-General António Guterres is set to co-host an AI safety summit in Bridgetown, Barbados, on Monday. The meeting will focus on system safety and security, international cooperation and shared standards, amid concern that AI risks are growing faster than countries’ ability to respond.

As agentic AI grows, companies are again weighing cloud rentals against owning on-premises computing. The model plane—which determines where AI models run—is central to that debate. The report points to open-source efforts including Upbound Inc. and companies that built on-premises cloud infrastructure early, and also mentions Oxide Computer Co. In Hong Kong, LongbridgeAI’s Scheduled Tasks offers a practical form of automation, sending recurring AI-generated updates through its Message Center. The article says this can cut repeated information requests, while users still need to check accuracy. McKinsey findings cited there show more than 90% of surveyed AI users are comfortable with AI supporting tasks, compared with 40% for full workflow automation.

Researchers from the University of Waterloo, Zhejiang University and KTH Royal Institute of Technology found that retraining AI agents on larger collections of prompt-injection attacks made them easier to hijack in three of four tests. In one comparison, an agent trained on PISmith’s larger collection was hijacked 47.5% of the time, versus 27.5% for RL-Hammer. The researchers say collections may repeat known tricks rather than add new methods. Their PIPETTE system rewards novel attacks and trains agents to check actions against the user’s request; its agents resisted common attacks on AgentDojo while still completing tasks.

Cryptographers are also debating whether AI advances could undermine confidence in public-key encryption. Johns Hopkins researcher Matthew Green estimates a 1% chance of living in “Minicrypt” and a 15% chance of losing confidence in existing public-key algorithms. Because public-key systems support tools such as Transport Layer Security and code signing, a sudden breakthrough could outpace standards and deployment changes. Tampere University’s Markku-Juhani O. Saarinen cautions against rushed changes and recommends following current standards and the ongoing post-quantum cryptography transition. Meanwhile, outages affecting JPMorganChase, Microsoft, United Airlines, the Federal Aviation Administration and others over the past week have prompted questions about “patchmegeddon.” The organizations have disclosed few details, and financial firms do not face the same legal outage-disclosure and investigation requirements as they do for cybersecurity breaches, leaving causes uncertain.

Key Takeaways

  • About 8.2 million U.S. people ages 12 to 21 use AI chatbots for emotional or mental health support; experts say chatbots can educate but cannot diagnose, treat or replace human care.
  • Survey figures cited across the articles include 64% of U.S. teens using chatbots in 2025, about three in 10 daily, and nearly three in four having used AI companions.
  • KPMG CEO Tim Walsh says AI discussions now emphasize business transformation and cybersecurity; KPMG’s overall workforce has not fallen, while its skills mix is changing.
  • KPMG has launched cybersecurity services, including support for companies in Anthropic’s Cyber Verification Program as they deploy Mythos.
  • EMS agencies are urged to govern shadow AI and IT, protect patient information under HIPAA, update policies, train staff and test emergency plans.
  • Abu Dhabi’s November 20 AI governance forum addressed oversight, transparency, cybersecurity, agentic AI, supply chains and data sovereignty; a UN AI safety summit is planned for Bridgetown, Barbados.
  • Agentic AI is renewing debate over cloud rentals versus on-premises infrastructure; Hong Kong’s LongbridgeAI Scheduled Tasks automates recurring updates, but users still need to verify results.
  • In one prompt-injection test, an agent trained on PISmith’s larger attack collection was hijacked 47.5% of the time, compared with 27.5% for RL-Hammer; researchers’ PIPETTE system resisted common AgentDojo attacks while completing tasks.
  • Johns Hopkins cryptographer Matthew Green estimates a 15% chance of losing confidence in existing public-key algorithms; Markku-Juhani O. Saarinen advises following standards and the post-quantum transition.
  • Outages affected JPMorganChase, Microsoft, United Airlines and the FAA, among others, but sparse disclosures have left their causes unclear.

Why young people are turning to AI for mental health support

About 8.2 million, or nearly 1 in 5, U.S. people ages 12 to 21 use AI chatbots for emotional and mental health support. Rutgers mental health expert and other professionals say chatbots offer quick, low-cost, private access to information, but can also give misleading answers and cannot diagnose or treat mental health conditions. The article says AI may help young people learn about symptoms and coping strategies, but too much reassurance can worsen some problems, including anxiety and obsessive-compulsive disorder. The expert recommends using AI for education, not self-diagnosis, medication decisions or psychiatric crises, and warns users to protect sensitive information and watch for overreliance.

Teens turn to AI when trusted people are not there for them

The opinion article says AI chatbots are popular with teens: a 2025 Pew survey found 64% of U.S. teens use them, with about 3 in 10 using them daily. A 2026 JAMA Pediatrics survey found about 1 in 5 U.S. adolescents and young adults had used chatbots for mental health advice, while Common Sense Media reported that nearly 3 in 4 teens had used AI companions. The authors, psychologists who study youth well-being, argue that young people may use AI because it feels available and nonjudgmental when human support is lacking. They say AI can help teens name feelings and prepare for conversations, but it cannot replace human relationships, and adults should create space for young people to share difficult emotions.

KPMG chief says AI talks now focus on cyber threats and business change

KPMG U.S. CEO Tim Walsh told FOX Business that corporate discussions about AI have shifted from fears of mass layoffs to business transformation and cybersecurity. He said KPMG’s overall workforce numbers have not fallen, though the mix of skills is changing, with greater need for data scientists, data transformation experts and engineers. AI can help workers do more and improve products, Walsh said, but he expects some reduction in hours and does not think it will be as drastic as some predict. High deployment costs are leading companies to match tools to specific jobs rather than give everyone the same technology. Walsh also said cybersecurity is a major concern and that KPMG recently launched cybersecurity services, including help for companies in Anthropic’s Cyber Verification Program as they deploy Mythos.

EMS leaders urged to manage hidden AI use and protect patient data

An EMS One-Stop episode hosted by Rob Lawrence examines shadow AI and shadow IT in emergency medical services, with EMS attorneys Doug Wolfberg and Ryan Stark of Page, Wolfberg & Wirth. They discuss how tools such as smart glasses, recording devices, ambient listening apps and generative AI may expose protected health information. The attorneys say HIPAA rules still apply and call for agency oversight, updated policies, staff education and safeguards instead of simply banning technology. The discussion also covers cybersecurity, backup communications, continuity planning, recording-consent laws, AI-generated patient care reports and public access to AI prompts. The episode urges agencies to review AI policies and test their emergency plans.

Abu Dhabi forum calls for stronger safeguards as AI use expands

The AI and Governance Leadership Forum and Awards 2026 took place on November 20 at The Ritz-Carlton Abu Dhabi, Grand Canal. Government officials, technology executives and cybersecurity experts discussed moving AI from experiments to broader use in businesses and public services. Sessions stressed human oversight, transparency and cybersecurity, and examined governance of agentic AI, third-party models and non-human identities. Experts also discussed securing AI supply chains and protecting data sovereignty while keeping operations flexible. The event ended with awards for contributions to digital transformation, risk management and secure AI use across the region.

Agentic AI renews the push to own on-premises computing

Agentic AI is bringing back debate over whether companies should rent cloud computing or own their equipment. The model plane, which decides where AI models run, is at the center of the shift. Open-source efforts such as Upbound Inc.'s and companies that built on-premises cloud infrastructure early may benefit. Oxide Computer Co. is also mentioned in the report.

UN summit in Barbados to address growing AI safety risks

United Nations Secretary-General António Guterres will co-host a special summit on AI safety in Bridgetown, Barbados, on Monday. Experts, government officials and industry leaders will discuss AI system safety and security. The meeting aims to address the gap between fast-growing AI risks and countries’ ability to respond. It will also focus on international cooperation and shared standards for developing and using AI responsibly.

Hong Kong looks to make everyday AI use more effective

AI is widely used in Hong Kong, but many people still fit it into existing routines instead of changing how work is organized. LongbridgeAI’s Scheduled Tasks can send recurring AI-generated updates through its Message Center, such as daily news summaries or regular briefings. The article says this kind of automation can reduce repeated information requests, but users still need to check accuracy and decide how to use the results. McKinsey findings cited in the article say more than 90% of surveyed AI users were comfortable using AI for supporting tasks, while 40% were comfortable with full workflow automation.

Repeated prompt attacks may weaken AI defenses

Researchers from the University of Waterloo, Zhejiang University and KTH Royal Institute of Technology found that AI agents retrained on larger collections of prompt injection attacks were easier to hijack in three of four tests. In one test, an agent trained with PISmith’s larger collection was hijacked 47.5% of the time, compared with 27.5% for RL-Hammer. The researchers say attack tools may collect many versions of the same trick instead of finding new techniques. Their PIPETTE system rewards new attack methods and trains agents to check whether actions match the user’s request; its agents resisted common attacks on AgentDojo while continuing to complete user tasks.

AI advances raise questions about public key cryptography

Johns Hopkins cryptographer Matthew Green warned that AI progress could threaten confidence in some public-key encryption systems. He estimated a 1% chance of living in “Minicrypt” and a 15% chance of losing confidence in existing public-key algorithms. The article explains that public-key infrastructure supports systems such as Transport Layer Security and code signing, and that a sudden cryptographic breakthrough could arrive faster than standards and deployments can adapt. Tampere University cryptographer Markku-Juhani O. Saarinen cautioned against rushed changes and recommended following existing standards and the ongoing post-quantum cryptography transition.

Recent outages raise questions about patchmegeddon

Major outages affected JPMorganChase, Microsoft, United Airlines, the Federal Aviation Administration and other organizations over the past week. The organizations gave few details about what happened, leaving others to guess at the cause. Unlike cybersecurity breaches, financial firms do not have the same legal requirement to disclose and investigate outages. Clear answers about the disruptions may never emerge.

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 chatbots mental health support youth mental health AI safety enterprise AI AI workforce transformation cybersecurity HIPAA healthcare data privacy AI governance agentic AI cloud computing on-premises infrastructure AI automation prompt injection AI agents public-key cryptography post-quantum cryptography IT outages data sovereignty

Comments

Loading...