Scary headlines about AI destroying humanity are drawing attention away from real problems happening right now, according to a global AI security leader. The expert urges organizations and the public to focus on immediate, tangible risks rather than speculative doomsday scenarios that dominate the news cycle.
Several major companies have learned hard lessons after rushing to adopt AI tools. Meta scaled back plans to cut staff and deploy more AI agents when employees found themselves spending too much time fixing errors. Walmart workers deal with unreasonable expectations and false alerts generated by AI systems. Microsoft discovered that training workers and pairing new engineers with experienced colleagues produced better results. Most executives now acknowledge that AI has not yet delivered meaningful productivity gains at their companies.
On the security front, researchers tested an AI agent that reviews cloud security findings and downgraded many high-severity alerts by examining the full environment rather than relying on static rules. The agent matched real cloud conditions 99.4 percent of the time in live probes focused on AWS environments. Meanwhile, Bloom Security uncovered a supply chain risk in npm Trusted Publishing tied to AI coding workflows, where AI agents install large numbers of packages daily, exposing projects to automated dependency attacks.
AI scams are also targeting crypto users with deepfakes and stolen data, including one scheme that defrauded victims of nearly $16 million. Chainalysis reports that scammers use AI tools to make attacks cheaper and more convincing, while wallet security improves by adding automated checks on risky transfers. In a separate development, AI excels at solving math problems but struggles with research judgment, according to Princeton's Pravesh Kothari and colleagues, who used AI to narrow the bounds of the Grothendieck constant in just weeks with human intervention only when models got stuck.
UNESCO launched four new Russian-language courses to help university teachers use generative AI responsibly, covering lecture preparation, prompt engineering, AI-assisted assessments and student safety guidance. Federal agencies looking to modernize are advised to use AI built into pre-certified software on government-approved cloud platforms to speed up adoption without sacrificing security. In China, AI spending is driving holiday sales, with foot traffic up 50 percent and AI appliance sales up over 40 percent as shoppers buy AI pet robots, glasses and 3D printers.
Key Takeaways
- <ul><li>AI security expert warns that fear of existential AI risks is overshadowing real, immediate threats</li></ul>
- <ul><li>Meta scaled back AI agent plans after employees spent too much time fixing errors</li></ul>
- <ul><li>Microsoft found that training and mentoring improved AI results more than going it alone</li></ul>
- <ul><li>Most executives report AI has not yet produced productivity gains at their companies</li></ul>
- <ul><li>AI cloud security agent downgraded many high-severity findings and matched real environments 99.4% of the time</li></ul>
- <ul><li>Bloom Security found npm Trusted Publishing supply chain risk tied to AI coding workflows</li></ul>
- <ul><li>AI crypto scams defrauded victims of nearly $16 million using deepfakes and stolen data</li></ul>
- <ul><li>Princeton researchers showed AI excels at math but needs human intervention for research judgment</li></ul>
- <ul><li>UNESCO launched four Russian-language AI courses for university teachers</li></ul>
- <ul><li>China saw AI appliance sales rise over 40% during National Day holiday as smart devices enter mass market</li></ul>
AI fear overshadows real threats says expert
A global AI security leader says scary headlines about AI ending humanity are hiding real threats that already exist. The expert believes people should focus on problems happening now. The article was updated within the last few hours. The report appeared on multiple news stations.
AI fear overshadows real threats says expert
A global AI security leader says scary headlines about AI ending humanity are hiding real threats that already exist. The expert believes people should focus on problems happening now. The article was updated within the last few hours. The report appeared on multiple news stations.
AI fear overshadows real threats says expert
A global AI security leader says scary headlines about AI ending humanity are hiding real threats that already exist. The expert believes people should focus on problems happening now. The article was updated within the last few hours. The report appeared on multiple news stations.
AI fear overshadows real threats says expert
A global AI security leader says scary headlines about AI ending humanity are hiding real threats that already exist. The expert believes people should focus on problems happening now. The article was updated within the last few hours. The report appeared on multiple news stations.
AI fear overshadows real threats says expert
A global AI security leader says scary headlines about AI ending humanity are hiding real threats that already exist. The expert believes people should focus on problems happening now. The article was updated within the last few hours. The report appeared on multiple news stations.
Companies scale back AI after rushed rollout causes errors
Many companies rushed to adopt AI tools but now face a flood of errors and low-quality work. Meta scaled back plans to cut staff and use more AI agents after employees spent too much time fixing mistakes. Walmart employees deal with unreasonable expectations and false alerts from AI tools. Microsoft found that training workers and pairing new engineers with experienced ones improved results. Most executives say AI has not yet produced productivity gains at their companies.
AI agent downgrades most high-severity cloud security findings
Researchers Leon Goldberg and Gal Engelberg tested an AI agent that reviews cloud security findings. The agent downgraded many high-severity alerts by checking the full environment instead of relying on static rules. The agent collects evidence from connected resources and other security products before assigning a grade. Live probes confirmed the agent's facts matched the real cloud environment 99.4 percent of the time. The study focused on AWS environments and read-only assessment, not autonomous action.
SCAI UCO building offers virtual online tour
SCAI UCO is a modern building that now has a virtual tour created by Visitas Virtuales Granada. The tour lets users explore the building's architecture, design and features from anywhere online. It gives visitors a detailed experience without needing to be physically present. The virtual tour is a way to see the space and learn about its innovative design.
AI excels at math but struggles with research judgment
AI models are very good at solving math puzzles but lack the judgment needed for research, according to Princeton's Pravesh Kothari and colleagues. The models were trained on solved problems but rarely learn the long process of mathematical discovery. Using AI, the team narrowed the bounds of the Grothendieck constant, a concept from 1953, in just weeks. Humans stepped in only when the models got stuck, around iteration 18. Mathematician Terence Tao argues that AI-generated proofs are often too large and hard for humans to understand.
UNESCO launches AI courses for university teachers
UNESCO launched four new Russian-language courses to help university teachers use generative AI responsibly. The courses cover preparing lectures and tests, prompt engineering, AI-assisted assessments and guiding students on safe AI use. The training is available online through UNESCO platforms. The initiative builds on UNESCO IITE's earlier work to strengthen teacher AI skills. A similar effort took place in September 2026.
AI in pre-certified apps speeds up government modernization
Federal agencies face delays adopting AI due to complex security rules. Hunter Downey of SAP NS2 says agencies should use AI built into pre-certified software on government-approved cloud platforms. He recommends focusing on secure data connections and both generative and agentic AI. This approach can improve security while speeding up modernization.
AI scams target crypto users as wallet security improves
AI tools make crypto scams cheaper and more convincing, targeting users rather than protocols. Chainalysis reports scammers use deepfakes and stolen data, including a scheme that defrauded victims of nearly $16 million. Security experts urge automated controls to filter risky transfers and approvals before users confirm them. Wallet security is catching up by adding checks independent of user decisions.
AI spending drives China's holiday economy
Chinese shoppers bought AI gadgets during the National Day holiday, showing smart devices are moving into the mass market. Stores sold AI pet robots, AI glasses, and 3D printers, with foot traffic up 50 percent and AI appliance sales up over 40 percent. China had over 6,000 AI enterprises and more than 140 humanoid robot developers in 2025. Government policy support is pushing AI-enabled consumption further.
Bloom Security finds supply chain risk in AI coding workflows
Bloom Security discovered a software supply chain risk in npm Trusted Publishing tied to AI coding workflows. AI agents install many packages daily, increasing exposure to automated dependency risks. A gap in repository identity verification could let attackers publish malicious packages under trusted names. Over 1,500 packages with the relevant configuration accounted for more than 10 million downloads.
Sources
- Will AI end humanity? Expert says fear overshadows real threats
- Will AI end humanity? Expert says fear overshadows real threats
- Will AI end humanity? Expert says fear overshadows real threats
- Will AI end humanity? Expert says fear overshadows real threats
- Will AI end humanity? Expert says fear overshadows real threats
- Companies rushed in on AI. Now some are pulling back
- Most high-severity cloud security findings are downgraded once an AI agent checks the environment
- SCAI UCO | Virtual tour generated by Visitas Virtuales Granada
- Princeton Engineering - AI is great at math, but does it know how to do research?
- UNESCO launches AI courses for university teachers
- How AI embedded in pre-certified applications can fast-track modernization
- AI Scams Target Crypto Users, Not Protocols, and Wallet Security Is Catching Up
- Science & Tech: AI-enabled spending fuels China's holiday ec
- Bloom Security Finds a Software Supply Chain Risk Hiding in AI Coding Workflows
Comments
Please log in to post a comment.