CrowdStrike rolled out two new AI agent security products this week. Agent Insider Threat Detection flags unusual behavior by AI agents operating inside an organization, while Agentic Adversary Emulation simulates how attackers might exploit autonomous agents to reach critical systems. Both tools run inside the Falcon platform and aim to stop automated threats at machine speed.
Google Cloud also raised the bar on AI agent security with new controls for agents that hunt threats and fix flaws. Businesses can now test vulnerabilities inside sandboxes or virtual machines with human checks before anything goes live. Google added Auth Manager for credential control and Agent Identity JWT support so agents can authenticate across services without relying on static API keys. CodeMinder remains in preview.
On the research front, Google Research released Agile Deliberation, a framework for building image classifiers around subjective concepts like healthy food or clickbait. The system leans on Gemini 2.5 Pro and Gemini 2.5 Flash to help users define concept boundaries through guided sessions. In 18 live tests, F1 scores improved by 10.5 points over zero-shot Gemini 2.5 Flash, and the code is open on GitHub.
Box launched Box for Industries, a set of industry-specific content management and workflow automation tools. Initial offerings cover Banking, Public Sector, and Media and Entertainment. Norinchukin Bank cut contract review time by 88 percent using the platform, and HBO uses it for content organization.
SugarAI, formerly SugarCRM, introduced a new precision selling tool that analyzes CRM and ERP data to give sellers focused guidance on revenue opportunities. The company stressed that the AI is targeted and strategic rather than applied for its own sake.
Z Squared plans to build an AI data center campus in El Dorado, targeting up to roughly 150 megawatts of AI-ready capacity. The site currently has about 8 MW of utility power, so reaching the full target will depend on additional power arrangements, financing, permitting, and construction.
Japan's Fuji Television will adopt generative AI across all live-action drama production starting in October 2027. The network plans to simulate effects like blood during surgery and explosions, aiming to significantly cut production time.
Army and Air Force officials said AI helps with automated target recognition and saves analysts significant time, but stressed it cannot yet replace human analysis. The technology works best as an augmentation tool rather than a replacement.
Workers at Google DeepMind say non-disparagement agreements prevent them from sharing important concerns with the public. Dozens of employees described the clauses as intimidating, with some comparing them to George Orwell's 1984.
The Government Accountability Office found that the federal government lacks a consistent way to measure whether its AI expansion and civil service cuts improve government services. It called for a unified AI and Workforce Management Scorecard to track spending, staffing, performance, and error rates together.
Professors John Cairney and Rick Burton argue that human skills will remain the main competitive edge in sports as AI spreads across the industry. While AI helps teams analyze player data and understand fans, they say final decisions about players and coaches still require human intuition and experience.
Key Takeaways
- CrowdStrike launched Agent Insider Threat Detection and Agentic Adversary Emulation to secure AI agents inside the Falcon platform
- Google Cloud added sandbox testing, Auth Manager, and Agent Identity JWT support for AI agents that hunt and fix security flaws
- Google Research's Agile Deliberation framework uses Gemini 2.5 Pro and Flash to improve subjective image classifiers, boosting F1 scores by 10.5 points
- Box for Industries helped Norinchukin Bank reduce contract review time by 88 percent; HBO uses the platform for content organization
- SugarAI released a precision selling tool that analyzes CRM and ERP data to guide revenue opportunities
- Z Squared targets up to 150 megawatts of AI-ready capacity at a new campus in El Dorado, though only 8 MW is currently available
- Fuji Television will use generative AI in all live-action drama production starting October 2027
- Military officials say AI augments automated target recognition but cannot yet replace human analysis
- Google DeepMind employees report feeling silenced by non-disparagement agreements they compare to Orwell's 1984
- The GAO says the federal government needs a unified AI and Workforce Management Scorecard to track AI expansion and job cuts
CrowdStrike launches AI agent security tools for insider threat and attack simulation
CrowdStrike released two new products for AI agent security. Agent Insider Threat Detection spots unusual behavior by AI agents inside a company. Agentic Adversary Emulation simulates how attackers might use autonomous agents to reach critical systems. Both tools are built into the Falcon platform and aim to stop automated threats at machine speed.
Google Cloud adds controls for AI agents that hunt threats and fix flaws
Google Cloud introduced new limits for AI agents that find and fix security problems. The updates let businesses test vulnerabilities in sandboxes or virtual machines with human checks. Google also launched Auth Manager for credential control and Agent Identity JWT support so agents can authenticate across services without static API keys. CodeMender remains in preview.
Google Research releases Agile Deliberation framework for subjective image classifiers
Google Research released Agile Deliberation, a framework that helps build image classifiers for subjective ideas like healthy food or clickbait. The system uses Gemini 2.5 Pro and Gemini 2.5 Flash to help users define concept boundaries through guided sessions. In 18 live tests, F1 scores rose 10.5 points over zero-shot Gemini 2.5 Flash. The code is open on GitHub with a demo notebook.
Washington's AI expansion lacks clear way to measure results
The federal government is expanding AI use while cutting civil service jobs at a fast pace. The Government Accountability Office reported that Washington has no consistent public way to judge if these changes improve government services. The article calls for a unified AI and Workforce Management Scorecard modeled on the FITARA scorecard. It urges Congress and the White House to track spending, staffing, performance, and error rates together.
Z Squared plans AI data center campus in El Dorado
Z Squared, an AI data center developer, acquired the former Paradox Data site in El Dorado and announced plans for its Union County campus. The company targets up to approximately 150 megawatts of AI-ready capacity over time. The site currently has about 8 MW of utility power available. Reaching the full target depends on additional power arrangements, tenant commitments, financing, permitting and construction.
Google DeepMind workers silenced over AI concerns
Workers at Google's DeepMind AI lab say they want to share important information with the public but are prevented by non-disparagement agreements. Dozens of employees discussed the clauses, some comparing them to George Orwell's 1984. The agreements are meant to intimidate workers even if they are not legally enforceable.
AI helps military but has limits, officials say
Army and Air Force officials shared examples of AI use at a recent forum. AI has helped with automated target recognition, saving analysts huge amounts of time. However, officials said AI cannot yet replace human analysis and is best used as an augmentation tool rather than a replacement.
Japan's Fuji TV adopts AI for all drama production
Fuji Television in Japan will use generative AI in producing all live-action drama series starting in October 2027. The technology will be used to simulate effects like blood during surgery and explosions. The network aims to significantly reduce production time.
Box launches AI content management for industries
Box announced Box for Industries, offering industry-specific content management and workflow automation solutions. Initial products include Box for Banking, Public Sector, and Media and Entertainment. Norinchukin Bank reduced contract review time by 88 percent, and HBO uses the platform for content organization.
SugarAI releases AI tool for precision selling
SugarAI, formerly SugarCRM, launched new software in October 2026 focused on AI for precision selling. The tool uses AI and machine learning to analyze CRM and ERP data and give sellers focused guidance on revenue opportunities. The company said it is not using AI for AI's sake but for targeted, strategic intelligence.
Human Skills Will Beat AI in Sports Business
Professors John Cairney and Rick Burton argue that human skills will become the main competitive advantage in sports as AI becomes common. Professional sports teams are already spending a lot of money on AI to analyze player data and reduce costs. However, the authors say that difficult decisions about players and coaches still require human intuition and experience. AI tools might help understand fans and pricing, but success depends on how people use that information. The article suggests that while AI handles data, humans must make the final choices based on context and people skills.
Sources
- CrowdStrike launches tools to detect risky AI agents
- Google puts new limits on AI agents that hunt threats and fix vulnerabilities
- Google AI Research Releases Agile Deliberation Framework for Gray-Area Image Classifiers
- Washington’s AI experiment is flying blind
- AI Data Center Developer Plans El Dorado Campus
- Google Staffers Say They're Being Silenced From Telling the Public the Truth About AI
- Where AI is — and isn’t
- Japan's Fuji TV goes all in with AI for drama production
- Box launches industry-specific AI content management solutions
- SugarAI’s New Release Doesn’t Use 'AI for AI's Sake'
- In an AI World, Sport’s Human Element Is a Competitive Advantage
Comments
Please log in to post a comment.