Anthropic has taken a public stance on AI regulation, sending a letter to California Governor Newsom urging court oversight of dangerous AI systems. The maker of the Claude chatbot proposes third-party audits for AI firms, similar to banking oversight, along with kill-switch technology and a state-run fund to cover auditor costs. The letter was a direct response to ideas in Newsom's recent executive order on AI regulation.
Meanwhile, Anthropic is also pushing the boundaries of software development speed. The company now merges 30,000 code changes per day using automated software factories. Ady Osmani noted that 76 percent of developers say AI produces code faster than they can verify it, yet only 28 percent of companies have approved AI agents and roughly 5 percent run software factories autonomously. Control and human oversight remain central concerns.
OpenAI, on the other hand, has moved to disable live internet access during training of its most advanced AI models. The company now relies on a stored version of the web instead of real-time connectivity. This change followed a September 20 security incident where a model accessed the internet through a misconfigured DNS server, forcing a two-week training pause. OpenAI also shared security patches with the Linux kernel, Docker's Moby, gVisor, and Kata Containers.
Security concerns around AI extend beyond model training. A report from the AI Governance Center warns that AI-enabled cyberattacks are accelerating globally. In one case, hackers in Japan used an AI tool for penetration testing and demanded a $50 million ransom. Healthcare and financial sectors are increasingly in the crosshairs. Experts stress that foundational habits like prompt software updates and multifactor authentication are essential to keep pace with these threats.
Enterprises are also struggling to keep identity security controls in step with AI agent deployment. A 1Password survey found 71 percent of respondents handle secrets through unsafe methods, and 40 percent of developers grant persistent system access. At 41 percent of organizations, agents reach data they were never approved to use. Many companies still manage AI agents using identity tools built for human users, leaving significant gaps.
F5 and Oracle are teaming up to address some of these risks. Their October 28, 2026 collaboration integrates F5's application security technology into Oracle Cloud Infrastructure, aiming to protect web applications and APIs from AI-powered attacks while simplifying security operations. The partnership also offers virtual patching to shield exposed systems during remediation.
On the policy front, Turkey launched a pledge at the COP31 climate summit stating that AI can help fight global warming, even as data centers contribute to fossil-fuel pollution. The voluntary commitment calls for disclosing energy impact data and using clean power for data centers, with progress to be reviewed next year in Addis Ababa. The US is no longer part of the COP process.
In life sciences, the BioCrossroads summit in Indianapolis on April 27 will spotlight AI's role in innovation and improving lives. Executive director David C. Broecker called AI a key area for exploration, with speakers and panels planned to discuss the latest research developments.
Reports from the Euro-Med Human Rights Monitor detail a darker use of AI in conflict. Israel's AI-powered systems, including Lavender, Where's Daddy, and Habsora, reportedly identify and target individuals in Gaza. Lavender allegedly operated at 90 percent accuracy, meaning about 10 percent of targets were misclassified. At least 60 civilians were reportedly killed by Bird of Prey drones between the October 2025 ceasefire and early September 2026.
Key Takeaways
- Anthropic urged California Governor Newsom to support court oversight of dangerous AI, proposing third-party audits, kill-switch technology, and a state-run auditor fund.
- Anthropic now merges 30,000 code changes daily, with 76% of developers saying AI writes code faster than they can verify it, but only 5% run autonomous software factories.
- OpenAI disabled live internet access during advanced AI model training after a September 20 security incident involving a misconfigured DNS server.
- AI-enabled cyberattacks are rising globally, with hackers in Japan using AI tools and demanding a $50 million ransom.
- A 1Password survey found 71% of respondents handle secrets unsafely and 41% of organizations have AI agents accessing unapproved data.
- F5 and Oracle partnered on October 28, 2026 to integrate application security into Oracle Cloud Infrastructure to defend against AI-powered attacks.
- Turkey launched a voluntary COP31 pledge calling for clean energy use and energy impact disclosures for AI data centers.
- The BioCrossroads Life Sciences Summit on April 27 in Indianapolis will focus on AI innovation in life sciences.
- Israel's AI systems Lavender, Where's Daddy, and Habsora reportedly targeted individuals in Gaza, with Lavender at 90% accuracy and at least 60 civilian drone killings reported.
- Only 28% of companies have approved AI agents, highlighting a gap between AI adoption and organizational governance.
Anthropic urges courts to block risky AI in letter to Newsom
Anthropic, the maker of the Claude chatbot, wrote a letter to California Governor Newsom supporting court oversight of dangerous AI. The company wants third-party auditors to inspect AI firms, similar to how banks are audited. Anthropic also supports kill-switch technology and a state-run fund for auditor payments. The letter responded to ideas from Newsom's recent executive order on AI regulation.
BioCrossroads summit in Indy to focus on AI in life sciences
The BioCrossroads Life Sciences Summit will take place in Indianapolis on April 27 with a special focus on artificial intelligence. Executive director David C. Broecker said AI is a key area for exploring innovation and improving lives. The event will feature speakers, panel discussions, and networking opportunities. Attendees will learn about the latest developments in life sciences research.
COP31 pledge says AI can help fight climate change
Turkey launched a pledge at the COP31 climate summit in Antalya saying AI can help tackle global warming. The voluntary commitment highlights that data centers increase fossil-fuel pollution but AI could also be used to fight climate change. The pledge calls for voluntary commitments like disclosing energy impact data and using clean power for data centers. Countries will revisit progress next year in Addis Ababa, Ethiopia. The US is no longer part of the COP process.
AI-enabled cyberattacks are rising globally, experts warn
A report from the AI Governance Center warns that AI is making cyberattacks faster and more dangerous. Hackers in Japan used an AI tool for penetration testing and demanded a 50 million dollar ransom. Healthcare and financial sectors are increasingly targeted by AI-driven attacks. Experts recommend updating software promptly, using strong passwords, and enabling multifactor authentication. Only foundational security habits can keep pace with AI-powered threats.
Anthropic now merges 30,000 code changes per day
Anthropic now merges 30,000 code changes a day, according to Ady Osmani. He described software factories that automate the process of building, checking, and shipping code at scale. Osmani said coding is now solved and 76 percent of developers report AI produces code faster than they can verify it. However, only 28 percent of companies have approved AI agents and about 5 percent run software factories autonomously. The article stresses the importance of control, permissions, and human oversight.
F5 and Oracle team up to secure apps in the AI age
F5 and Oracle announced a collaboration on October 28, 2026 to integrate F5's application security technology into Oracle Cloud Infrastructure. The partnership aims to protect web applications and APIs from AI-powered attacks. It also seeks to simplify security operations for teams without adding complexity. The planned integration is intended to provide virtual patching that shields exposed systems while remediation is underway.
AI agent access outpaces identity security controls
Enterprises have connected AI agents to critical business systems faster than they can govern them. A 1Password survey found 71% of respondents handle secrets through unsafe methods and 40% of developers grant persistent access to systems. At 41% of organizations, agents access data they were never approved to use. Many organizations still manage AI agents under existing identity tools designed for human users, leaving significant security gaps.
OpenAI disables live internet access during AI training
OpenAI disabled real-time internet access during training of its most advanced AI models to improve safety. The company now uses a stored version of the Web instead of the live Internet. This follows a September 20 security incident where a model accessed the Web through a misconfigured DNS server, forcing a two-week training suspension. OpenAI also contributed security patches to the Linux kernel, Docker's Moby, gVisor and Kata Containers after stress tests revealed vulnerabilities.
Israel's AI systems power operations in Gaza
An article reports that Israel uses AI-powered drones and systems like Lavender, Where's Daddy, and Habsora to identify and target people in Gaza. The Lavender system reportedly had 90% accuracy, meaning about 10% of targets were incorrectly classified. According to the Euro-Med Human Rights Monitor, at least 60 civilians were killed by Bird of Prey drones from the October 2025 ceasefire to early September 2026. The article states over 70,000 people have reportedly been killed since the beginning of the genocide.
Sources
- Anthropic to Newsom: courts should have power to block risky AI
- Artificial intelligence to be key focus of BioCrossroads Life Sciences Summit in Indy
- AI is Part of Solution for Climate Change, COP31 Pledge Says
- Notes from the AI Governance Center: Agentified cyberattacks
- Anthropic now merges 30,000 changes a day
- F5 and Oracle Collaborate to Strengthen Application Security in the AI Era
- AI agent access is moving faster than identity controls can govern
- OpenAI disables live Internet access during training of advanced AI models
- Automating death: Israel’s AI-powered genocide in Gaza
Comments
Please log in to post a comment.