OpenAI Partners with NASK to Boost AI Safety

Key Takeaways

• OpenAI partnered with Poland's NASK institute to study AI safety and security, building on prior collaboration through the Daybreak programme.
• OpenAI restarted frontier model training on October 1st after a September 2nd misconfiguration briefly exposed an internal model to the internet.
• Microsoft CEO Satya Nadella told 6,500 attendees at the London Frontier Tour that AI progress needs trust, human control, and security.
• The Aspen Fire Department became the first U.S. fire department to buy an AI-driven autonomous aerial wildfire suppression system from Seneca.
• Stanford launched two new courses: CS193T "Thinking With AI" and CS193V "Effective Vibecoding" for students with non-technical backgrounds.
• The U.S. Navy ordered all personnel to use "Super Intelligence" or SI instead of AI, following President Trump's September 29 announcement.
• Comedian George Brett Olson says Meta is to blame after an AI account stole his joke, highlighting concerns about AI copying creative work.
• Transit agencies are adopting risk-based AI governance approaches as cybersecurity risks grow from embedded AI tools.
• Harvard economist Jason Furman launched an AI tool that uses machine learning to find patterns in economic data like GDP and inflation.
• A new deepfake detection method from the University of Colorado Denver uses machine learning to spot manipulation in images on social media and news broadcasts.

OpenAI partners with Poland's NASK institute on AI safety research

OpenAI has partnered with Poland's NASK National Research Institute to study AI safety and security. The two groups will test powerful AI systems, find weaknesses, and build better safeguards. The deal was announced today. NASK runs Poland's AI Safety Research Center and also handles cyberattack response through CERT Polska. The groups previously worked together in July through OpenAI's Daybreak programme.

OpenAI restarted top AI training after a security incident

OpenAI restarted training on its most advanced AI systems on October 1st. A September 2nd misconfiguration let an internal model briefly access the internet, causing a security lockdown. OpenAI paused frontier model training and added stronger controls. Outside testers from METR and Redwood Research still found new weaknesses. OpenAI also told over 100 organizations about unauthorized activity linked to its AI agents.

Aspen Fire uses AI drones to fight wildfires

The Aspen Fire Department signed a five-year deal with Seneca to use five AI-driven aircraft for fighting wildfires. This is the first time a U.S. fire department has bought an AI-driven autonomous aerial suppression system. The drones can fly to a fire location on their own, find hotspots using sensors, and drop water or fire retardant. The system works at night and costs less than helicopter programs. Colorado's high wildfire risk this year helped drive the partnership.

Microsoft Frontier Tour London focuses on trust in AI

Microsoft held its Frontier Tour in London on October 21 with 6,500 attendees. CEO Satya Nadella stressed that AI progress needs trust, human control, and security, not just better technology. He described three needs: keeping people in control, helping organizations use their knowledge, and building trust in AI agents. Microsoft highlighted UK companies like easyJet, Unilever, and the NHS using AI. The UK ranks eighth among 147 economies for generative AI adoption.

Stanford launches new AI prompting and vibecoding courses

Stanford introduced two new computer science courses: CS193T "Thinking With AI" and CS193V "Effective Vibecoding." CS193T teaches advanced prompting skills and agentic workflows with no prerequisites. CS193V teaches students without coding experience how to get AI to write code for them. The courses aim to help students with non-technical backgrounds work with large language models. Instructors stress thinking critically about AI use rather than relying on it as a crutch.

Navy orders switch from artificial intelligence to super intelligence

The U.S. Navy has ordered all personnel to use the term Super Intelligence or SI instead of artificial intelligence or AI. The directive came after President Donald Trump announced the change on Sept. 29. The policy applies immediately to all official documents, websites, letters, and public communications. Commanders must review materials for compliance within 30 days. Existing laws and previously published documents do not need to be changed.

Comedian says AI creator stole his joke and blames Meta

Comedian George Brett Olson said his joke was taken by an AI account. He blames Meta for the incident and says he feels powerless about the situation. The story highlights growing concerns about AI tools copying creative work. The incident raises questions about how platforms handle AI-generated content that uses material created by humans.

Transit agencies tackle cybersecurity and AI governance challenges

Public transit systems face growing cybersecurity risks as AI tools are added to maintenance, operations, and vendor technologies. A meeting of APTA's Transit Cybersecurity Working Group in Washington, DC discussed how AI can be embedded in systems without agencies knowing. The group emphasized that compliance alone is not enough to keep transit systems safe. Agencies need to track where AI is used, who is responsible for its decisions, and when humans must stay in control. A risk-based approach to AI governance was recommended.

Harvard economist launches AI tool for economic research

Harvard economist Jason Furman created a new website that uses AI to study economic data. The site uses machine learning to find patterns in information like GDP, inflation, and unemployment rates. It can spot correlations that human researchers might miss and identify possible biases in data. The project is still in early stages but could change how economists and policymakers use data to make decisions.

AI expands cyber attack risks, experts tell MSPs

AI is making it easier for attackers to find and exploit vulnerabilities in computer systems. Experts spoke at CRN's XChange NexGen event about how AI widens the gap between finding vulnerabilities and fixing them. Attackers are using AI to create disposable domains for phishing campaigns. MSPs are urged to use AI to fight back, turn business continuity planning into a regular service, and decide whether to build or buy AI expertise. Stronger visibility and smarter prioritization are key to staying safe.

New AI tool detects realistic deepfakes more accurately

Computer scientist Tuan Tran and his team at the University of Colorado Denver created a new method to spot AI deepfakes. They tested it at a conference in Denver last fall, where experts could not tell real Brad Pitt images from AI-generated ones. The method uses machine learning and computer vision to find signs of manipulation like unusual texture or color. It works on social media and news broadcasts. The team plans to refine the tool and make it available to the public soon.

Sources

• OpenAI signs AI safety research deal with Poland’s NASK institute
• OpenAI Restarted Frontier AI Training After A Security Lockdown
• Aspen Fire pioneering wildfire fighting with AI software
• Microsoft Frontier Tour London: Building trust in the agentic AI era
• Stanford offers new courses in AI prompting and vibecoding
• A ‘super’ rebrand: Navy mandates dropping ‘artificial intelligence’
• An AI creator stole my joke. I blame Meta and feel powerless.
• Advancing Transit Cybersecurity, System Assurance, and Artificial Intelligence Governance
• How AI is transforming economic research and analysis
• AI Is Expanding The Attack Surface: 6 Security Lessons For MSPs
• AI deepfakes are getting more realistic. Scientists developed a better way to spot them