Experts agree on seven principles to address AI-related risks

Key Takeaways

Experts Gather to Address AI Risks in Nuclear and Biological Threats

More than 100 experts met at California's Asilomar Conference Grounds to discuss AI's impact on nuclear and biological threats. The meeting launched the Asilomar Process to develop practical safeguards for AI-related risks. AI companies are restricting access to advanced models due to concerns about misuse. Seven principles were agreed upon to address the challenges. These principles aim to protect human survival and ensure AI systems do not exacerbate global catastrophic risks.

AI Changing Biological and Nuclear Risks, Governance Must Adapt

AI is changing the landscape of biological and nuclear threats, requiring new governance approaches. The James Martin Center for Nonproliferation Studies convened experts to discuss AI's impact on these threats. Seven principles were developed to guide AI applications in nuclear and biological security. These principles prioritize human survival and meaningful human control over AI systems. The AI industry must work with security experts to address these challenges.

BlackFog Launches ADX Vision for macOS Security

BlackFog has launched ADX Vision for macOS, extending its shadow AI security platform to Apple endpoints. This provides enterprises with a single AI data-loss policy across Windows and macOS devices. ADX Vision offers direct visibility into AI-bound data flows on macOS devices, addressing a significant blind spot in enterprise AI governance.

Strongest AI Teams Will Use Diverse Models

The strongest teams of AI agents will be built using different models. Research shows that diverse agentic systems outperform homogeneous ones. AI companies are increasingly using agentic AI, but without attention to diversity, promised benefits may not materialize. A study demonstrated significant performance improvements from diversity in agentic systems.

Anjney Midha on AI's Shift from FLOPs to Megawatts

Anjney Midha, founder of AMP PBC, discusses the AI industry's shift from focusing on computational power (FLOPs) to energy efficiency (megawatts). This transition is driven by growing concerns about AI's environmental impact and energy consumption. Midha emphasizes the importance of aligning organizational culture with stated goals and executing strategies practically.

AI's Assault on the Black Body: A New Consumer Protection Crisis

AI-powered 'nudify' apps are generating fake intimate images, causing harm, particularly for Black women. This technology industrializes historical exploitation, treating bodies as commodities. Experts advocate for a proactive regulatory framework to govern AI before harm occurs. Current laws and voluntary efforts are insufficient to address the issue.

Business AI Agents Become Big Tech's Next Battleground

Big tech firms are investing heavily in AI agents for customer service, aiming to convert attention into transactions. Meta, Google, Amazon, Microsoft, and OpenAI are competing in this space. AI agents offer capabilities like answering questions, remembering product details, and responding in multiple languages. However, they also raise concerns about data control and commercial value.

Healthcare IT Stories: AI Security, Workforce Gaps, and Tech Layoffs

Healthcare IT faces several challenges, including AI security issues, workforce gaps, and tech layoffs. AI-driven layoffs are affecting the industry, with Oracle's largest-ever layoff being a recent example. Remote patient monitoring (RPM) studies have raised doubts about its effectiveness as a standalone intervention. Cybersecurity coordination is becoming a growing obligation for health systems.

Virginia Law Permits AI Cameras for Traffic Enforcement

A new Virginia law allows local governments to use AI-powered cameras to enforce traffic rules for pedestrian safety. The technology, developed by Obvio, can issue citations to drivers who fail to stop for pedestrians or disregard stop signs. The law aims to reduce traffic deaths and improve safety.

The Future of AI: Small, Cheap, and Unprofitable

A recent study suggests that small language models running on desktop computers may soon be able to handle most tasks currently performed by large language models. This trend could undermine the business case for large AI models and data centers. Small models are improving rapidly in terms of intelligence per watt, making them more cost-effective.

Top AI Bots on the Internet

AI bots now account for 57.5% of all HTTP traffic, with Googlebot dominating at 27.8%. Meta-ExternalAgent, ClaudeBot, Bytespider, and GPTBot are among the top AI bots. These bots collect training data, power search and retrieval, and act on behalf of users. Cloudflare tracks and categorizes bot activity by crawl purpose.

Sources

• AI Is Changing Biological and Nuclear Risks; Governance Must Change Accordingly
• AI is changing biological and nuclear risks; governance must change accordingly
• BlackFog Launches ADX Vision for macOS - Extending Shadow AI Security to the Enterprise
• The Strongest Teams of AI Agents Will be Built Using Different Models
• Anjney Midha on AI's FLOPs to Megawatts Transition
• AI's Assault On The Black Body :The New Consumer Protection Crisis
• More than chatbots: why business AI agents are Big Tech’s next product battleground
• Healthcare IT's defining stories: AI security, workforce gaps, RPM limits, and tech layoffs
• Virginia law permits AI cameras to enforce traffic rules for pedestrian safety
• The future of AI may be small, cheap and unprofitable
• These Are The Top AI Bots On The Internet [June 2026]