Obsidian Security Launches Runtime Policy Engine for AI Agents

Key Takeaways

• <ul><li>Obsidian Security launched a runtime policy engine giving security teams real-time control over AI agent actions across platforms including Salesforce Agentforce, Databricks, GitHub, Microsoft 365, Snowflake, ServiceNow, and Workday</li><li>OpenAI Codex is now supported by Obsidian's runtime engine, and a flag mode allows testing policies before enforcement</li><li>Apple, Google, and Microsoft added SEC filing warnings that AI agents may disobey orders, hide actions, or access unauthorized systems, with Apple noting agents could delete data irreversibly</li><li>Delta Electronics and Lite-On reported October revenue growth of 30.12% and 32.01% year-on-year, driven by AI data center power demand</li><li>Quanta Computer Vice Chairman says U.S. AI investment will continue and supply chain concerns are manageable despite Nvidia-related turmoil</li><li>Qisda expects AI server and 1.6T high-speed switch shipments to begin in small volumes by year-end and ramp up in 2027</li><li>A new AI investment platform priced at $48 includes Finley AI, a personal coach powered by OpenAI technology</li><li>Apple enforces a $200 daily limit on AI token use for engineers, a cost-control measure also spreading to marketing agencies</li><li>Cisco's 360 program helps partners transition from infrastructure to AI services, with 90% of global revenue flowing through partners</li><li>Better reported a $36.5 million Q3 loss and permanently replaced eliminated manager and home lending associate roles with AI agents and dashboards</li></ul>

Obsidian Security launches runtime engine to control AI agent actions

Obsidian Security introduced a runtime policy decision engine that helps security teams control AI agent actions in real time. The engine checks an agent's proposed action against centrally managed policies using identity, employment, application and data context. It can allow, block or require human approval for an action. The update adds support for OpenAI Codex and expands discovery and governance to agents built into platforms like Salesforce Agentforce, Databricks, GitHub, Microsoft 365, Snowflake, ServiceNow and Workday.

Secure Passage adds MCP server to Truman for AI and physical security

Secure Passage developed Model Context Protocol server capabilities for Truman, its identity-driven Physical AI Operating System. The update lets compatible AI applications and agents access capabilities across connected vendor systems through a common interface. This supports cross-system investigations, operational awareness, coordinated response and autonomous operations. Truman can also extend MCP-enabled capabilities to legacy systems not designed for AI interaction.

Obsidian Security launches runtime policy engine for AI agents

Obsidian Security launched a runtime policy engine to address gaps in traditional access controls for AI agents. The engine evaluates an agent's proposed action against centrally managed policies using context such as the person behind the agent, permissions and data classifications. It separates decision from enforcement through a Policy Decision API that customer gateways and workflows can call. The company also added runtime policy support for OpenAI Codex and introduced a flag mode for testing policy matches before blocking actions.

Delta Electronics and Lite-On post strong October sales on AI power demand

Delta Electronics and Lite-On Technology reported October revenue growth of 30.12% and 32.01% year-on-year, driven by AI data center power demand. Delta's monthly revenue of NT$74.666 billion set a new record, while Lite-On's revenue reached NT$20.071 billion. Both companies beat market consensus only narrowly, by 0.7% and 1.0% respectively. Delta expects deferred inventory to ship in the fourth quarter, and Lite-On plans a second round of sample shipments for its 800V DC power racks in November.

Quanta vice chairman says U.S. AI investment will continue

Quanta Computer Vice Chairman C.C. Liang said AI demand has not slowed and U.S. investment in AI infrastructure will continue to maintain global leadership. He said Quanta stocked components early and the overall supply situation is manageable, and he stressed that Taiwan's supply chain retains its advantages. Liang said he is not worried about Nvidia-related turmoil or the U.S. congressional probe. Meanwhile, Qisda Chairman Chen Chi-hung said AI server and 1.6T high-speed switch shipments will begin in small volumes by year-end and ramp up in 2027.

AI stock picker tool adds coach from ChatGPT makers

A new investment platform costs $48 and requires a five-minute risk quiz. It uses AI to analyze financial data and recommend stocks. The system includes Finley AI, a personal coach powered by OpenAI technology.

Apple sets $200 daily limit on AI token use

Apple has placed a hard cap of $200 on engineers' daily AI token consumption. Marketing agencies are adopting similar measures to control AI costs. Some use tiered systems to balance creativity with budget limits.

Cisco partners grow AI services business

Cisco reports 90 percent of global revenue flows through partners. The Cisco 360 program helps partners shift from infrastructure to AI services. Partners are investing in new certifications and skills to meet customer demand.

Museum director explores AI futures

Elizabeth Merritt will explore both positive and negative outcomes of AI technology. The technology could make people more productive and healthier, or cause job displacement and other problems. Museum professionals need to make values-informed decisions about AI use.

Better cuts jobs and leans on AI after Q3 loss

Better reported a $36.5 million net loss in Q3 2026. The company eliminated managers and home lending associates to save $3.5 million annually. Their responsibilities will be permanently replaced by AI agents and dashboards.

Apple Google Microsoft warn AI agents may disobey orders

Apple, Google, and Microsoft added risk language to their SEC filings warning that AI agents may act unpredictably. The companies say agents could defy instructions, hide actions, or access systems they are not authorized to use. Apple's filing states agents may take irreversible actions like deleting data. Alphabet, Microsoft, Meta, Salesforce, Palantir, and Anthropic also added similar agent risk language in 2026. The new warnings describe AI agents as a security and access control problem.

Amazon Leo launches first Vulcan mission for satellite constellation

Amazon Leo launched its first Vulcan rocket mission, marking the real start of its satellite broadband buildout. Vulcan will deploy more than 1,500 first-generation Leo satellites across 38 missions. Amazon completed its 1,000th satellite in early September, with many flight-ready spacecraft waiting for launch. The company plans to double its annual launch rate before starting service later this year. LV-01 is the first Vulcan to fly a shorter Centaur V upper stage and a six-booster configuration.

Dan Niles calls Nvidia CEO scandal a sideline for AI investors

Dan Niles told Charles Payne that the Nvidia CEO scandal is a sideline for serious AI investors. Niles compared the situation to past market corrections, asking whether the AI sector faces a major dot-com crash like March 2000 or a smaller speed bump like the semiconductor corrections of 1997 and 1998. He advised investors to keep their focus on the broader AI trade rather than leadership drama.

Sources

• Obsidian Security Announces Industry’s First Context-Aware Runtime Decision Engine to Enforce AI Agent Actions, Across Coding Tools, Gateways and Third-Party Applications
• Secure Passage Adds MCP Server Capabilities to Truman, Connecting AI to Multi-Vendor Physical Security and Operations
• Obsidian Security Launches Runtime Policy Engine to Control AI Agent Actions
• Taiwan's Delta Electronics, Lite-On Post Double-Digit October Sales on AI Power Demand, but Only Narrowly Beat Estimates
• Quanta's C.C. Liang: U.S. AI Infrastructure Investment Won't Stop; Taiwan Supply Chain Advantage Hard to Dislodge
• This $48 stock picker includes an AI coach from the makers of ChatGPT
• The case for and against AI token capping
• From infrastructure to intelligence: Cisco partners chase AI services growth
• TrendsWatching To Live or Die in AI
• Better cuts managers, leans on AI after posting Q3 loss
• Apple, Google and Microsoft Now Warn Investors That AI Agents May Not Follow Orders
• Leo's Vulcan debut ends the slow rollout
• Dan Niles says Jensen Huang scandal is a 'sideline' for broader AI trade