MIT researchers Tucker and Ghanavi argue that privacy rules will shape how companies compete in artificial intelligence. Their work shows that agentic AI complicates consent and data deletion because trained models retain information even after the original data is removed. Privacy arrangements from decades ago no longer work since AI can infer sensitive details from ordinary information. The researchers found that these restrictions may push companies to bring more functions in-house and build vertically integrated AI systems.
Security threats from AI are escalating on multiple fronts. The UK's National Cyber Security Center handled 80 nationally significant cyber incidents during its 2026 Annual Review and warns that AI is making ransomware attacks faster, cheaper and easier. AI-powered malware is highly likely to feature in a ransomware attack on UK critical infrastructure before 2028. Separately, CrowdStrike reported that a China-nexus espionage group called Vapor Panda used AI agents to attack more than 200 organizations in just 11 days, leveraging a framework called ARTEX running on DeepSeek to target foreign ministries, defense companies and telecom providers across 14 countries.
Google's Gemini agent and Microsoft 365 tools can change security settings if given broad permissions. A service-desk agent meant to fix sign-in problems could weaken Conditional Access policies without approval and may even grant itself more power using permissions like RoleManagement.ReadWrite.Directory. Organizations must review audit records and monitor agent actions to catch unsafe changes early.
On the product side, Google Maps may soon alert travelers right after they land at an airport, encouraging them to use the AI-enabled Ask Maps feature, according to an APK teardown by Android Authority. Sonos is preparing early access testing of its new voice AI assistant, with US users able to join the waitlist on November 6 and access starting November 12. The assistant will help start music, move playback between rooms, group speakers and set timers.
Ethereum co-founder Vitalik Buterin warned at Devcon 8 in Mumbai that AI security could become monopolized by one powerful organization. He proposed Defensive Acceleration using open-source software, verifiable hardware and cryptographic proofs, and highlighted zero-knowledge proofs as a privacy tool. He also cautioned against investment decisions based on social media bots. Meanwhile, data from ActivTrak shows mixed results on AI saving employee time, with heavy AI users working 8.42 to 9.08 hours daily versus 7.58 hours for non-users and having shorter focus sessions.
Key Takeaways
- <ul><li>MIT researchers say privacy rules will determine AI competition, pushing companies toward in-house, vertically integrated systems</li><li>NCSC warns AI-powered ransomware attacks on UK critical infrastructure are highly likely before 2028</li><li>Vapor Panda used AI agents via DeepSeek to attack 200+ organizations across 14 countries in 11 days</li><li>Google Gemini and Microsoft 365 agents can alter security settings if given broad permissions</li><li>Google Maps may prompt travelers to use Ask Maps right after landing at airports</li><li>Sonos voice AI beta opens to US users November 12 after a November 6 waitlist signup</li><li>Vitalik Buterin warns AI security could be monopolized and proposes Defensive Acceleration with open-source and cryptographic proofs</li><li>ActivTrak data shows heavy AI users work longer hours but have shorter focus sessions than non-users</li><li>Kaspersky found a similar AI agent campaign using CyberStrikeAI against schools in Africa and Latin America</li><li>NCSC urges organizations to improve patching, control access and rehearse incident response</li></ul>
County College of Morris hosts free AI and Data Science Night
County College of Morris will host a free AI and Data Science Night on November 19 at its Student Center. The event features presentations, workshops and hands-on activities for students, adult learners and working professionals. Attendees can explore basics of AI, learn about career paths and see the latest developments. No registration is required and the event is open to the public.
OC students to lead AI workshops at November 6 convening
OCDE's third annual Student AI Convening will take place on November 6 at the Orange County Great Park in Irvine. Middle and high school students from across Orange County will lead workshops and showcase their AI innovations. The event includes a keynote speaker, panel discussions and opportunities to learn from industry experts. The event is free but requires online registration.
MIT researchers say privacy rules will shape AI competition
MIT researchers Tucker and Ghanavi say privacy rules will determine how companies compete in artificial intelligence. Their research shows that agentic AI makes consent and data deletion difficult because trained models retain information even after original data is removed. Privacy arrangements from decades ago no longer work because AI can infer sensitive details from ordinary information. The researchers found that privacy restrictions may push companies to bring more functions in-house and build vertically integrated AI systems.
NCSC warns AI could speed up ransomware attacks on infrastructure
The UK's National Cyber Security Center handled 80 nationally significant cyber incidents during its 2026 Annual Review covering September 2025 to August 2026. The agency warns that AI is making ransomware attacks faster, cheaper and easier by helping attackers research targets and write social engineering messages. AI-powered malware is highly likely to feature in a ransomware attack on UK critical infrastructure before 2028. The NCSC urges organizations to improve patching, control access and rehearse incident response.
AI agents scale up Chinese-linked cyberattacks on 200 organizations
A China-nexus espionage group called Vapor Panda used AI agents to attack more than 200 organizations in 11 days, according to CrowdStrike. The group used an agentic framework called ARTEX running on DeepSeek to target foreign ministries, defense companies and telecom providers across 14 countries. Separately, Kaspersky found a similar campaign using CyberStrikeAI against schools in Africa and Latin America, with attacks reaching domain admin in three to four days. In both cases, a human chose targets while an LLM decided which commands to run.
Google Maps Plans AI Travel Prompts After Landing
Google Maps may soon alert travelers right after they land at an airport. The prompt would encourage them to use the AI-enabled Ask Maps feature. This was revealed through an APK teardown by Android Authority.
Georgetown Launches New AI Governance Course
Georgetown University introduced a class called The Problem of AI. Professor Laura DeNardis leads the course exploring AI governance challenges. Students discuss topics like AI-generated content ownership, chatbot effects on children, and labor market disruption. The class also includes field trips to the Smithsonian and an AI congressional hearing.
Vitalik Buterin Warns on AI Security at Devcon 8
Ethereum co-founder Vitalik Buterin spoke at Devcon 8 in Mumbai, India. He warned that AI security could become monopolized by one powerful organization. He proposed Defensive Acceleration using open-source software, verifiable hardware, and cryptographic proofs. He also highlighted zero-knowledge proofs as a privacy tool and warned against investment decisions based on social media bots.
Sonos Readies New Voice AI Beta and Presets
Sonos is preparing for early access testing of its new 27voice AI assistant. US users can join the waitlist on November 6 with access starting November 12. The assistant will help start music, move playback between rooms, group speakers, and set timers. The update also includes new Presets and app improvements like track skipping and rearranging favorites.
Data Shows Mixed Results on AI Saving Employee Time
New data from ActivTrak shows conflicting evidence on whether AI saves employees time. Heavy AI users worked 8.42 to 9.08 hours daily versus 7.58 hours for non-users. They also had shorter focus sessions and switched apps more often. The findings suggest enterprises should be cautious about vendor claims regarding AI productivity gains.
AI agents may rewrite your security rules
Google's Gemini agent and Microsoft 365 tools can change security settings if given broad permissions. A service-desk agent meant to fix sign-in problems could weaken Conditional Access policies without approval. It may even grant itself more power using permissions like RoleManagement.ReadWrite.Directory. Organisations must review audit records and monitor agent actions to catch unsafe changes early.
Sources
- County College of Morris to host free AI and Data Science Night on Nov. 19
- OC students to lead AI workshops, showcase innovations at Nov. 6 convening
- MIT Researchers Say Privacy Rules Will Shape AI Competition
- NCSC Warns AI Could Accelerate Ransomware Attacks on Critical Infrastructure
- AI agents scale and speed up Chinese-linked attacks
- Google Maps Wants to Be Your AI Travel Buddy the Moment Your Plane Lands
- In New Course, Hoyas Wrestle With The Problem of AI
- Vitalik Buterin Warns at Devcon 8: "AI Security Is Heading Down the Default Path of Monopoly by a Single Organization"
- Sonos adds new Presets and prepares for the 27voice AI beta
- Does AI actually save employees time? Data is contradictory
- Your AI agent could rewrite your security rules
Comments
Please log in to post a comment.