CrowdStrike has launched SafeMind, an AI security suite built with Nvidia to combat AI-based cyberattacks. The system uses two models, Red Tempest for attack path detection and Blue Solano for remediation, operating within a digital twin of a customer's environment. At the Fal.Con event, CEO George Kurtz revealed that attacker breakout time has dropped from minutes to seconds, with the fastest recorded attack taking just 27 seconds. CrowdStrike also introduced Project QuiltWorks, a cross-industry alliance focused on identifying and fixing AI-related security vulnerabilities.
On the policy front, AI's role in designing DNA, proteins, and drugs is raising new biosecurity concerns. The White House released policy under Executive Order 14292 in July 2026, and companies like OpenAI and Anthropic have released safety frameworks. A June 2026 letter urged Congress to mandate screening on synthesizers, with bills like H.R. 3029, S. 3741, and S. 4363 under consideration. Meanwhile, cybersecurity experts say major AI companies, including those led by Sam Altman and Dario Amodei, are excluding outside security professionals from safety planning despite public commitments to slow the race toward more powerful AI systems.
In other developments, a CAST policy brief highlights how AI is already embedded in farming through seed selection, autonomous weed spraying, and agricultural robotics. Author Alex Thomasson of Mississippi State University says federal policy is lagging behind real-world adoption, calling for better rural broadband and clearer data ownership rules. Separately, Microsoft has set an October 7 event focused on local AI and future Windows PCs, with CEO Satya Nadella and NVIDIA CEO Jensen Huang in attendance. The event is expected to cover NVIDIA RTX Spark, Surface hardware, and AI processing for consumer computers.
NIST and CISA have released token security guidance that leaves AI agent risks largely unaddressed, noting that prompt injection can steer agents with valid tokens toward unauthorized actions. Senator Richard Blumenthal has called for AI oversight modeled on FDA pharmaceutical regulation, warning that delayed action could lead to losing control over autonomous rogue bots. In the private sector, Latham & Watkins is investing in AI by purchasing Nvidia servers and partnering with PacerPro, which works with legal AI platform Harvey, while also creating new workflows and guidelines for daily AI use.
Key Takeaways
- <ul><li>CrowdStrike and Nvidia launched SafeMind, featuring Red Tempest and Blue Solano models, to counter AI-based cyberattacks with breakout times as fast as 27 seconds</li><li>George Kurtz revealed at Fal.Con that attacker breakout time has shrunk from minutes to seconds, per CrowdStrike's 2026 Global Threat Report</li><li>Project QuiltWorks was introduced as a cross-industry alliance to address AI-driven security vulnerabilities</li><li>OpenAI and Anthropic have released biosecurity frameworks amid concerns over AI-designed DNA, proteins, and drugs, with Congress weighing bills H.R. 3029, S. 3741, and S. 4363</li><li>Cybersecurity experts say OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei are excluding outside security professionals from AI safety planning</li><li>Google DeepMind and General Intuition are using AI-generated games and 3D environments to train robots and autonomous systems for real-world tasks</li><li>NIST and CISA token guidance does not cover AI agent risks, leaving enterprises vulnerable to prompt injection attacks via valid tokens</li><li>Senator Blumenthal proposed bipartisan AI oversight legislation modeled on FDA pharmaceutical regulation to prevent loss of control over autonomous systems</li><li>Microsoft scheduled an October 7 event with Satya Nadella and Jensen Huang to showcase local AI, NVIDIA RTX Spark, and Surface hardware for Windows PCs</li><li>Latham & Watkins invested in Nvidia servers and partnered with PacerPro and Harvey to build AI tools and establish new firm-wide AI guidelines</li></ul>
CrowdStrike launches SafeMind AI to counter fast cyberattacks
CrowdStrike introduced SafeMind, an AI security suite built with Nvidia to fight AI-based cyberattacks. It uses two models, Red Tempest and Blue Solano, to find and fix attack paths in a digital twin of a customer's environment. Research shows AI attackers are extremely fast, with the fastest attack taking just 27 seconds. CrowdStrike also launched Project QuiltWorks, a cross-industry alliance to find and fix AI-related security vulnerabilities.
CrowdStrike reveals AI threat response at Fal.Con event
At the Fal.Con event, CrowdStrike CEO George Kurtz highlighted that attacker breakout time has shrunk from minutes to seconds. The company's 2026 Global Threat Report found an average eCrime breakout time of 29 minutes, with the fastest at 27 seconds. CrowdStrike partnered with Nvidia to release SafeMind, featuring Red Tempest for attack path detection and Blue Solano for remediation. The company also emphasized ecosystem collaboration through Project QuiltWorks to address AI-driven vulnerabilities.
AI and biosecurity raise new policy concerns in the US
AI is being used to design DNA, proteins, and drugs, raising biosecurity concerns about whether current safety rules are enough. In July 2026, the White House released policy under Executive Order 14292 to replace earlier dual-use and DNA-screening rules. Companies like OpenAI and Anthropic have released frameworks, and a June 2026 letter asked Congress to mandate screening on synthesizers. Congress may consider bills like H.R. 3029, S. 3741, and S. 4363 to address these emerging risks.
CAST report urges policy support for AI in farming
A new CAST policy brief examines how AI is transforming crop production through seed selection, autonomous weed spraying, and agricultural robotics. Author Alex Thomasson of Mississippi State University says AI is now embedded in farming practice, not just research. The report calls for better rural broadband, data ownership clarity, workforce development, and continued research funding. It warns that federal policy is not keeping up with AI adoption in agriculture.
Cybersecurity experts say AI companies exclude them from safety plans
Cybersecurity experts say major AI companies are leaving them out of safety planning. OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei have publicly agreed that the race to build more powerful AI systems needs to be slowed. The article reports on concerns that outside security professionals are not being included in efforts to address risks from advanced AI development.
AI-generated games are training the next generation of robots
AI-generated games are now being used to train robots and AI systems for real-world tasks. Companies like General Intuition use gamer clips to create datasets that teach AI how to solve problems. Google DeepMind creates 3D environments where AI can learn safely. These virtual lessons help autonomous cars and kitchen robots make better decisions. Billions of hours of gameplay are building the skills needed for physical AI.
NIST-CISA token guidance leaves AI agent risks unaddressed
New guidance from NIST and CISA on securing identity tokens does not cover risks from AI agents. The guidelines recommend securing signed tokens but acknowledge that AI agents create additional access challenges. Experts say enterprises need to track who provisions agent credentials and set expiration dates. Prompt injection can steer agents with valid tokens toward unauthorized actions. Organizations are urged to use audience restrictions and shared-signal mechanisms to limit damage from compromised tokens.
Senator Blumenthal calls for AI oversight before it is too late
Senator Richard Blumenthal says new AI products should face objective review before release. He introduced bipartisan legislation to regulate AI while balancing safety and innovation. He compared AI oversight to the FDA's pharmaceutical regulation model. Blumenthal warned that delayed action could lead to losing control over autonomous rogue bots. He cited social media as an example of failing to regulate harmful technology in time.
Microsoft sets October 7 event for local AI and Windows PCs
Microsoft scheduled an October 7 event to explore local AI and future Windows PCs. CEO Satya Nadella, NVIDIA CEO Jensen Huang, and Microsoft executive Pavan Davuluri will attend. The event will cover Windows, NVIDIA RTX Spark, Surface hardware, and the broader PC ecosystem. It may reveal more details about the Surface Laptop Ultra. The focus is on putting AI processing power inside consumer computers for faster, offline AI features.
China sells AI tokens but carriers should build agent stores
China's telecom carriers are selling AI token plans as consumer packages. Analyst Suman Kanuganti says carriers are acting as a distribution layer rather than an aggregation layer. He argues carriers could own the full stack by building an agent store. A personal agent tied to a phone number could expand revenue by about $10 per subscriber per month. Kanuganti outlines three revenue layers including native AI use cases, LLM aggregation, and agent-to-agent transactions.
Latham Embraces AI With New Servers and Guidelines
Latham & Watkins is investing heavily in artificial intelligence. The firm bought servers from Nvidia to build AI tools and protect client data. It also partners with PacerPro, which works with legal AI platform Harvey. Latham has created new workflows and guidelines for using AI in its daily work.
Sources
- AI-based cyberattacks reshape cybersecurity
- CrowdStrike unveils AI defenses against rapidly evolving cyber threats
- Artificial Intelligence And Biosecurity Issues
- AI in Agriculture Brief Highlights Crop Production
- Cybersecurity experts say AI giants are shutting them out of safety plans
- AI-generated games — How Video Games Are Now Training Next-Gen AI and Robots
- AI agent authorization risks remain a gap in new NIST-CISA token security guidance
- Sen. Blumenthal urges AI oversight: 'We're on the verge of losing control'
- Microsoft sets 7 October event to explore local AI and future Windows PCs
- Suman Kanuganti: China Is Selling AI Tokens, but Carriers Should Build the Agent Store
- How AI Is Changing Latham, From Servers to Workflow Guidelines
Comments
Please log in to post a comment.