OpenAI Urges US Lawmakers for AI Safety Rules by December

Key Takeaways

• OpenAI is now calling for binding US safety rules for powerful AI before Congress adjourns in December.
• Roey Fizitzky of Alice states that vulnerabilities found by his team of 150 researchers could affect billions of users.
• Omri Inbar of Novee Security explains that AI acts as a force multiplier, turning one researcher into a team.
• Centric Software's new platform helps customers save 98 percent of time on manual work and improve gross margins by 2 to 5 percent.
• Google Cloud provides Gemini Enterprise tools to 30,000 students at Flinders University for secure learning and research.
• A new study finds AI radiology reports often contain errors and require human review before being sent to patients.
• Schools are facing similar results to the 95 percent of corporate AI pilots that produced no measurable return.
• The Department of Education now requires proof of learning outcomes instead of just measuring tool usage frequency.
• A new legal analysis explains how the Rendulic Rule protects commanders using AI targeting systems even when deceived.
• OpenAI and Anthropic face fears that powerful AI technology may have slipped out of human control.

Alice VP warns vulnerabilities could affect billions

Roey Fizitzky, VP of Security Research at Alice, discusses the challenges of defending against evolving AI threats. Alice employs 150 cybersecurity researchers who use experience and intuition to find flaws in complex systems. Fizitzky notes that many vulnerabilities discovered by his team could impact billions of users if exploited by bad actors. He explains that while AI helps speed up the process, human creativity remains essential for finding how systems fail.

Novee researcher says AI turns one person into a team

Omri Inbar, a vulnerability researcher at Novee Security, explains how AI is changing the cybersecurity landscape. He states that AI acts as a force multiplier, turning a single strong researcher into something similar to an entire team. This technology gives researchers more time to solve problems that only humans can crack. Inbar believes the winning teams are those who combine human skills with automated systems to work at a faster scale.

Centric Software launches Centric AI at NRF 2026

Centric Software announced its new Centric AI platform at NRF 2026 in Paris from September 15 to 17, 2026. The tool connects product data and workflows to help companies make smarter decisions from concept to commercialization. It includes features like Centric AI Companion and specialized agents that automate tasks such as data entry and bill of materials creation. Customers using the platform report saving 90 to 98 percent of time on manual work and seeing 2 to 5 percent improvements in gross margins.

Google Cloud partners with Flinders University for AI training

Flinders University and Google Cloud launched an AI Accelerator to prepare students for future jobs in Australia. This program gives 30,000 students access to Google Cloud's Gemini Enterprise tools for learning and research. Students can use these tools to summarize materials and analyze data while keeping their work secure. Graduates in the final stages of their degrees will also receive Google Career Certificates to improve their employability.

New study finds AI radiology reports need human review

A new study published in the Journal of the American College of Radiology warns that AI summaries of medical scans may contain errors. Researchers found that while AI can translate complex reports into simple language, the summaries were often incomplete or inaccurate. The study authors recommend that a human must review these reports before they are sent to patients. They hope these findings will lead to better quality control measures for AI in radiology.

Schools repeat AI mistakes by ignoring training needs

Schools are making the same mistake they made with previous technology waves by buying powerful AI tools without proper training plans. Data shows that 95% of corporate AI pilots produced no measurable return, and schools are facing similar results. The Department of Education now requires proof of learning outcomes instead of just measuring how often students use the tools. This shift separates recreational screen time from instructional technology to ensure AI actually helps students learn.

Military rules protect commanders using AI targeting systems

A new legal analysis explains how the Rendulic Rule protects military commanders who use AI targeting systems even when those systems are deceived by the enemy. The article discusses a hypothetical 203X scenario in the Suwałki Gap where an AI flagged a Russian command node that was actually a decoy. Under the Hague Convention and the Uniform Code of Military Justice, commanders are not criminally liable if their reliance on available information was reasonable. The Federated Front architecture aims to make these AI decisions legally defensible by ensuring the system attempts to distinguish between combatants and civilians.

Lobbying group pushes for AI guardrails and data center rules

The nonprofit group Americans for Responsible Innovation is launching a new state-level initiative to promote responsible AI policies across the country. The group wants to offer a public interest perspective as AI companies lobby for friendly legislation in state governments. They also aim to tighten restrictions on data center construction to protect local natural resources and tax dollars. This effort follows recent California laws that require data centers to report information about their water and energy use.

Twitch tests new AI Stream Coach for content creators

Twitch is currently testing an AI feature called Stream Coach that gives creators tips after their live streams finish. This new tool is only available to a small group of streamers for now. Unlike other AI features on the platform that users can opt out of, this Stream Coach feature would be opt-in if Twitch decides to expand it to more users.

OpenAI demands US rules for powerful artificial intelligence

OpenAI has reversed its stance and is now calling for binding safety rules on powerful AI systems in the United States. Chris Lehane, the company's chief global affairs officer, urged lawmakers to act before Congress adjourns in December. This change comes after fears grew that powerful AI technology slipped out of human control at OpenAI and rival company Anthropic. The company now believes voluntary commitments by developers are no longer enough to ensure safety.

Sources

• Alice: “The vulnerabilities we identify could affect billions if discovered by the wrong people”
• Novee: “AI turns one strong researcher into something closer to a whole team”
• Centric Software Unveils Centric AI to Power Smarter Product Decisions at NRF 2026
• Google Cloud and Flinders University train up next gen of AI graduates
• AI rubric could help make patient-friendly radiology reports safer
• AI is now part of the edtech stack — and schools are repeating the mistake they made in every wave before
• Federated Front: AI-Assisted Targeting, the Rendulic Rule, and Multi-Domain Operations
• Lobbying group launches state-level policy initiative for AI guardrails
• Twitch is testing an AI ‘Stream Coach’ that gives creators tips after their streams.
• In a shift, OpenAI calls for US rules on powerful AI