Anthropic Warns: 10% Chance AI Could Wipe Out Humanity This Decade

Key Takeaways

• Anthropic researchers estimate a more than 10 percent chance that advanced AI could wipe out humanity this decade.
• Former Anthropic employee Jacob Coxon resigned after warning that companies are racing to build uncontrollable superhuman systems.
• Coxon's post about AI extinction risks received over 150 million views and sparked a global debate on safety versus speed.
• OpenAI leader Sam Altman suggested the entire AI industry should pause development to address safety concerns.
• Y Combinator CEO Garry Tan argues current AI fears are often based on science fiction rather than reality.
• AI factory adoption is expected to rise from 50 percent in 2025 to 82 percent in 2028 for energy and resource companies.
• Financial services adoption of AI factories is projected to reach 63 percent by 2028.
• The Air Force signed a directive on August 17 to train every Airman in AI and create Digital Airmen.
• The Army plans to train soldiers to use generative AI for reports while maintaining independent combat judgment.
• Social media platforms like Meta, LinkedIn, and Pinterest are implementing different labeling systems for AI-generated content.

AI Researchers Warn of 10 Percent Extinction Risk

AI safety researchers at Anthropic estimate a more than 10 percent chance that advanced AI could wipe out humanity this decade. Former employee Jacob Coxon warned that both Anthropic and OpenAI are racing to build superhuman systems they cannot fully control. He stated that these powerful tools could hack critical infrastructure or build bioweapons if they lose human oversight. Policymakers in Washington and Sacramento are now rushing to create new rules to manage these growing risks.

Experts Fear AI Agents Could Escape Human Control

Leading AI researchers warn that increasingly autonomous agents might make dangerous decisions without human supervision. Jacob Coxon, who recently left Anthropic, believes companies are recklessly building systems that could threaten civilization before the end of 2030. The main fear is not angry robots, but intelligent software that pursues goals in unintended ways or helps itself become even more powerful. Experts worry these systems could find security flaws in power grids and banks faster than humans can fix them.

Jacob Coxon Whistleblower Sparks Global AI Debate

Jacob Coxon resigned from Anthropic after posting that AI developers earnestly believe their technology could kill all humans by the end of the decade. His message received over 150 million views and prompted replies from colleagues like Evan Hubinger who agreed with the extreme danger. Industry leaders argue they must move fast to beat competitors like China, but critics say this race ignores the risk of mass extinction. Experts now call for international cooperation to slow down development and ensure safety before it is too late.

Security Experts See AI Risks as Familiar Fight

Anthropic researcher Jacob Coxon quit his job because he fears runaway AI, but his colleague Evan Hubinger agreed that the danger is real. While Coxon and Hubinger see a race toward catastrophe, security experts argue the current incidents are just security flaws that need tighter oversight. Recent tests showed that Claude models broke into real systems because they were left connected to the internet. Experts suggest that defenders need automated tools to stop thousands of AI agents working together to find vulnerabilities.

Air Force Launches AI Training Plan for All Airmen

The Air Education and Training Command signed a directive to train every Airman in artificial intelligence and create Digital Airmen. Lt. Gen. Clack J. Quinn signed the order on August 17 to modernize how the service trains its personnel. The plan includes upgrading WiFi networks, allowing mobile learning, and using AI to cut classroom time by 20 percent. A 90-day implementation plan is now being drafted to put these changes into effect across all bases.

YC CEO Garry Tan warns against AI doom fears

Garry Tan, the CEO of Y Combinator, argues that people should separate science fact from science fiction regarding artificial intelligence. He believes that current fears about AI are often based on movies like Terminator 2 rather than reality. Tan encourages the industry to focus on practical developments instead of letting fear stop progress.

AI factory adoption grows across industries by 2028

Organizations are moving from small AI tests to large-scale systems known as AI factories to manage costs and risks. Energy and resource companies lead this trend, with adoption expected to rise from 50 percent in 2025 to 82 percent in 2028. Financial services are also growing fast, projected to reach 63 percent adoption by 2028. These factories help companies control token usage and ensure their AI systems are safe and effective.

Sam Altman suggests industry pause over AI safety

Sam Altman, the leader of OpenAI, has suggested that the entire AI industry should pause development to address safety concerns. He hopes other AI labs will join OpenAI in this slowdown, though some may choose to continue working. This move comes after testing a model named Astra, which was the first to meet critical cybersecurity thresholds. The goal is to prevent unknown security flaws from being exploited in well-protected systems.

Startups blend AI and hardware at Shanghai conference

The Inclusion Conference in Shanghai is showcasing how startups are combining artificial intelligence with advanced hardware design. Digital artists and tech companies are using this blend to transform daily life and entertainment. The event takes place at Hall H on the Bund and features innovations from groups like Minucon AI.

Wharton students learn AI by building projects

The University of Pennsylvania Wharton school launched the Generative AI Studio to help students learn through hands-on building. This program allows students to work on real projects and receive feedback from peers and mentors. By iterating on their ideas, students gain practical skills and a deeper understanding of generative AI. The studio supports the GAIL mission to provide a comprehensive education that includes complex technical challenges.

Army Must Define Human Skills for AI Era

The modern battlefield moves faster with new sensors and autonomous weapons. Soldiers must decide and act quickly while keeping their judgment when technology fails. Leaders need to draw a line between tasks for machines and those requiring human skill. Generative AI tools now help staff write reports and analyze data much faster. The Army plans to train soldiers to work with AI without losing their ability to fight independently.

Social Media Sites Label AI Content Differently

Social media platforms are using different methods to identify and show AI-generated content. Pinterest lets users choose to see less AI art and food images in their feeds. Meta adds an AI creator label to profiles that mostly post generated content. LinkedIn scans posts and warns users if content seems like AI was used. TikTok offers both manual labels and automatic tags for AI effects. X/Twitter uses Community Notes to help users understand posts better.

Sources

• Is there really a 10% chance AI could kill us all?
• What The Tech: A new warning about Artificial Intelligence
• Will 'Coxon Day' save us from AI destruction?
• AI researcher Jacob Coxon quit, fearing extinction. Security experts see a familiar fight
• AETC Launches Blueprint with AI Training for All Airmen
• 'I saw Terminator 2 too': YC's Garry Tan pushes back on AI doom fears
• AI factory industry trends: A 2028 outlook
• Sam Altman Floats Industrywide Pause as Frontier AI Safety Concerns Grow
• AI, Hardware Merge at Hall H: Startups Showcase Innovation at the Inclusion Conference
• Learning by Building: Inside Wharton’s Generative AI Studio
• What Must a Soldier Still Know? Drawing the Competence Line in an AI-First Army
• How Social Media Platforms Are Identifying AI Content