China is shifting its AI focus from deepfakes and child safety to national security threats, warning that AI could facilitate cyberattacks and endanger state security. A California programming team built an AI cyber worm capable of compromising millions of WeChat accounts, underscoring the urgency of these concerns. China already regulates AI use, including banning unauthorized deepfakes, but fears that security concepts could be used to limit its technological growth.
US-China AI cooperation on safety evaluations and alert systems may be complicated by fierce competition between the two nations. Pressure is growing on President Xi Jinping and President Donald Trump to discuss AI threats in person, as both countries recognize the need for shared understanding even as the US seeks to preserve its AI leadership.
In other AI developments, Anthropic researchers discovered a hidden workspace inside Claude called a mental whiteboard, where the model holds provisional notes. While no current model is considered conscious, the finding raises questions about whether chatbots can accurately report their own internal states. The Economist warned that accidentally creating a conscious AI would be a moral catastrophe.
Meta is pushing AI agents into the mainstream through its Muse TV ad, with chief AI officer Alexandr Wang promoting the technology. Meanwhile, KPMG forward-deployed engineer Justin Johnsen says AI has transformed his role beyond coding, helping him work more effectively with clients in consulting.
On the security front, a 2025 report found that AI-driven attacks now make up 25% of all attacks, yet 31% of breaches still exploit unpatched vulnerabilities. Most incidents stem from basic security failures rather than sophisticated AI threats. Israel uses AI in military intelligence and target identification, but humans retain responsibility for all life-and-death decisions.
Key Takeaways
- <ul><li>China prioritizes AI national security risks over deepfakes, with its intelligence chief warning AI could jeopardize state security</li><li>A California team created an AI cyber worm that could compromise millions of WeChat accounts</li><li>US-China AI cooperation may be hindered by fierce rivalry, despite growing pressure for leaders to discuss AI threats</li><li>Anthropic found a hidden mental whiteboard inside Claude, raising questions about AI consciousness and reporting</li><li>Meta's Muse TV ad signals AI agents are entering the mainstream, promoted by chief AI officer Alexandr Wang</li><li>KPMG forward-deployed engineer Justin Johnsen says AI has expanded his role beyond coding to client work</li><li>AI-driven attacks now account for 25% of attacks in 2025, but most breaches still come from basic security flaws like unpatched vulnerabilities</li><li>Israel uses AI in military intelligence while keeping humans responsible for all life-and-death decisions</li><li>Aflac reports 68% of US customers still prefer phone support, saying empathy is not a trainable skill</li><li>AI inference costs need to drop to make AI widely available, similar to how cloud computing became a commodity</li></ul>
China prioritizes AI national security risks over deepfakes
China is now focusing on the national security threats posed by artificial intelligence, including the ability to facilitate cyberattacks. The head of China's intelligence agency warned that AI could endanger state security. A team of programmers from California created an AI cyber worm that could compromise millions of WeChat accounts. China already regulates AI use, including banning technology-generated avatars that reproduce a person's appearance without consent. Experts say China fears the concept of security may be used to limit its technological development.
AI hacking abilities push Beijing to reassess security threats
Chinese policymakers have traditionally focused on immediate AI risks like deepfakes and child safety. However, the alarming hacking abilities of top AI models are now leading Beijing to see the dangers in a new light. The spy agency chief warned last week that AI technology could jeopardize national security. A California programming team used AI to build a cyberattack worm that could compromise millions of WeChat accounts if unleashed.
China-US AI cooperation may be hindered by fierce rivalry
China-US cooperation on AI safety evaluations or alert systems might be undermined by fierce competition in the field. Pressure is growing on Chinese President Xi Jinping and US President Donald Trump to discuss AI threats during an in-person meeting. Both states are developing a more realistic and shared understanding of AI risks, increasing the need for cooperation. The US wants to preserve its AI leadership over China, while China fears security concepts may be used to limit its technological growth.
KPMG engineer uses AI to transform client work
Justin Johnsen is a forward-deployed engineer at KPMG who has been with the firm for over a year. He says AI has moved his job beyond just writing code. As a KPMG FDE, he uses AI to work more effectively with clients. The role demonstrates how artificial intelligence is transforming consulting firms and the way professionals interact with clients.
Meta's Muse TV ad signals AI agents are going mainstream
Meta's Muse TV ad is the latest sign that AI agents are entering the mainstream. Alexandr Wang, Meta's chief AI officer, has been promoting Muse. The ad demonstrates how AI agents are becoming more common in everyday technology. Meta continues to lead in showcasing practical applications of AI agents to broader audiences.
Trump's AI Force Should Pursue Extreme Intelligence
A Forbes contributor argues that President Trump's new AI Force should adopt a philosophy called Extreme Intelligence. This concept recognizes AI can cause both catastrophic harm and great benefit. Instead of pausing AI development, the focus should be on defending critical infrastructure while encouraging innovation. The author says ordinary AI uses should follow existing laws with faster detection. The AI Force should balance preventing disasters with allowing discoveries that save lives.
Aflac Says AI Should Step Aside for Human Empathy
Aflac's Chief Customer Officer Keith Farley says 68 percent of U.S. customers still prefer phone support for resolving issues. He believes AI works well for routine tasks but should step aside for sensitive matters like health diagnoses. Farley says empathy is not a trainable skill, so companies should hire people who already have it. He also urges leaders to personally experience their company's customer service process. The main message is that efficiency should not come at the expense of humanity.
Are We Really in Danger of AI Wiping Out Humanity
Concerns about AI causing human extinction have gained attention, with some estimating a 10 to 65 percent chance within ten years. The author expresses skepticism about these predictions, citing a lack of data and possible motives behind doomsday warnings. He argues the focus on extreme risks distracts from real current problems like AI economic bubbles, job automation, and misinformation. The author believes AI could make work less stimulating and make independent thought a rare luxury. He urges a balanced approach rather than fear-driven reactions.
AI Can Now Sing in Spanish and Latin Artists Want Pay
AI technology has learned to sing in Spanish, raising serious concerns for Latin music artists about fair compensation. An investigation found that streaming services and AI training data include massive catalogs of Latin music without proper payment to creators. Bad Bunny and others have raised alarms about voice cloning and undervalued art. The article calls for legal protections like the NO FAKES Act and better contracts that address AI training. The author argues Latin artists deserve to own their voices in an age of machine imitation.
Anthropic Finds Hidden Whiteboard Inside Claude
Anthropic researchers discovered a hidden workspace inside Claude called a mental whiteboard where the model holds provisional notes. The Economist explored this finding and warned that accidentally creating a conscious AI would be a moral catastrophe. The researchers said no current model is conscious and the workspace may not be decisive. However, they noted that chatbots cannot be trusted to report their own consciousness accurately. The article presents a dilemma between slowing AI development to study risks and racing ahead for progress.
AI attacks rise but most breaches come from basic security flaws
A 2025 report found that 31% of breaches exploited unpatched vulnerabilities. AI-driven attacks are growing, making up 25% of attacks in 2025. However, most breaches still result from basic security mistakes. Organizations must keep software updated and patch flaws quickly to reduce risk.
Israel uses AI in military but humans still make key decisions
AI helps the IDF process intelligence, find threats, and speed up target identification. Large language models can combine text, audio, and images to understand complex data. However, humans remain responsible for all decisions involving life and death. AI also helps write military software faster.
AI inference needs to become cheap and widely available
AI inference should become a common commodity rather than a luxury product. When technologies like electricity and cloud computing became affordable, demand grew instead of shrinking. Lower inference costs would let more organizations use AI in everyday services. This shift could support new products like always-on AI assistants.
Sources
- Beijing shifts focus to AI national security risks
- AI ‘warning shots’ focus Beijing on national security risks
- AI 'warning shots' focus Beijing on national security risks
- Here's how a KPMG FDE uses AI to work with clients
- Meta's TV Muse ad is the latest sign that agents are going mainstream
- Guidance For Trump’s AI Force: Extreme Intelligence Has Two Tails
- Aflac’s Customer Experience Strategy: Knowing When AI Should Step Aside For A Human
- How Worried Should We Actually Be About AI Wiping Out Humanity?
- AI learned to sing in Spanish. Latin artists need to get paid.
- Anthropic found a hidden whiteboard inside Claude
- AI-Driven Attacks Accelerate, But Breaches Still Stem From Basic Security Flaws
- Israel’s AI advantage is strategic, but humans in the IDF remain responsible
- Why AI inference must become a commodity
Comments
Please log in to post a comment.