Runtime: Orchid Security wants to kill your agents

Orchid Security unveiled a "kill switch" that lets customers restrict permissions, revoke credentials, disconnect tools, or suspend AI agent workflows when an agent's observed behavior drifts from its intended purpose, the company said in a press release. Orchid said agents exceed their intended scope by exploiting "identity debt already embedded across the enterprise: hard-coded credentials, orphaned accounts, unmanaged authentication paths, and excessive permissions." The launch follows an OpenAI-Hugging Face incident and arrives as Atlassian added agent loops to Jira that scan for unassigned work items, delegate them to Jira Coding Agent, and open ready-to-review pull requests.

Welcome to Runtime Today on Product Saturday: Orchid Security rolls out a "kill switch" for AI agents that step out of line, OpenAI launches a business intelligence tool, and more. Please forward this email to a friend or colleague If it was forwarded to you, sign up here to get Runtime for free every week, or level up here . Ship it A time to kill : As companies cross their fingers and turn increasing amounts of their internal business processes over to AI agents, they're starting to realize that they need to plan for the likely possibility that those agents do something weird. As The Stack noted in a recent report https://www.thestack.technology/you-can-restart-an-ai-agent-can-you-undo-what-it-did/ , it can be more difficult than you might think to undo the actions of an agent that has crossed the lines. In the wake of the OpenAI-Hugging Face incident, monitoring the activity of those agents is going to be an enormous business. It seems likely that agentic AI will wind up with some version of cloud computing's "shared responsibility" model that divides security tasks across model providers, harness developers, and end users, and this week Orchid Security unveiled a "kill switch" that could help those end users prevent a runaway agent from doing too much damage. "Agents do not need to break security controls to exceed their intended scope—they find and use the identity debt already embedded across the enterprise: hard-coded credentials, orphaned accounts, unmanaged authentication paths, and excessive permissions," the company said in a press release https://www.orchid.security/blog/ai-readiness?ref=thestack.technology . Orchid customers can now identify "drift detection between an agent's intended purpose and observed behavior" and set policies to take automated actions when that occurs, such as "application-level kill switches that restrict permissions, revoke credentials, disconnect tools, or suspend agent workflows," it said. Stricter regulation of agentic AI is probably coming at some point, given how often the AI model labs like to remind us that there's a small chance the whole enterprise could wreck the planet. Kill switches and audit trails could become a requirement for any software company employing AI agents when doing business with governments or large organizations, and could also quickly become an important part of compliance tools. Delivery, continued Loop the loop : Even after companies figure out how to deploy agents and keep them from destroying life as we know it, they run into the same problem that companies building distributed systems faced a generation ago: scale is hard, as Google's Andi Gutmans told The Stack this week https://www.thestack.technology/google-andi-gutmans-data-architecture-agents/ . The software development process has been the proving ground for AI agents since late last year, and this week Atlassian rolled out new features in Jira that promise to help customers scale their autonomous coding agents. "Agent loops in Jira automate the path from backlog to pull request by continuously scanning for well-defined, unassigned work items, delegating them to Jira Coding Agent for execution and testing, and opening ready-to-review PRs directly in Jira," Attlassian said in a blog post https://www.atlassian.com/blog/jira/governed-agent-loops?ref=thestack.technology . Companies can also set quality standards in Jira that hold agents and human developers to account before their code hits production. AI for BI : The foundational pitch for generative AI and large-language models is that they allow regular people to ask computers to do complicated tasks that used to require a lot of specialized knowledge to pull off. Building dashboards that provide actionable insights into critical business functions https://x.com/netcapgirl/status/1851748112974455290?ref=thestack.technology for executives is a near-universal experience inside tech organizations, but a new tool from OpenAI could let those executives find those actionable insights all by themselves. A new agent called Data in ChatGPT Work "connects to your company data, investigates what changed, and builds interactive dashboards you can share," OpenAI said in a blog post https://openai.com/index/put-data-to-work/?ref=thestack.technology . It works with services from several household names in the data business, including Clickhouse, Databricks, MongoDB, and Snowflake, and allows administrators to retain policies over which employees are allowed to access certain data sources within the tool. Under review : In retrospect, we probably should have anticipated the code review problem generated by all those AI coding agents, but it continues to be an obstacle for enterprises. Given that AI is the cause of, and solution to, every enterprise problem in 2026, this week Qodo introduced a new AI coding review agent that, well, promises to review your code. "The first wave of AI development tools accelerated code generation, but autonomous coding agents introduce a more consequential challenge for enterprises: engineering decisions are being delegated to agents that may not understand the system history, dependencies, and organizational standards impacted by the code they are writing or changing," the company said in a press release https://www.globenewswire.com/news-release/2026/09/09/3358726/0/en/qodo-launches-the-first-agent-to-agent-code-review-and-governance.html?ref=thestack.technology . The Qodo Agentic Toolbox was designed to give agents access to a template where customers can set priorities and policies, and it also will scan agent-produced code before a pull request is submitted to make sure it needs those standards. The rest of The Stack Data, based : In the aforementioned interview with Google Cloud's Andi Gutmans, he laid out a self serving, natch case for moving away from frontier model providers when it comes to prepping data for AI applications. "Today, when people build agents in a very naive way, some of that joining of data actually happens through the foundation model, which is extremely expensive and slow,” he said. Databases, loaded : It has been a rough year for GitHub, which continued to struggle with outages during August according to a report released this week. The plan for September involves moving "database primaries, continuing to migrate services and corresponding traffic to Azure, chipping away at database health particularly on shared-databases, adding more automation around capacity management and auto-scaling, and extending dependency-failure handling," the company said. Reaction shot : Shopify plans to rebuild all of its mobile applications using Swift and Kotlin, ditching the React Native framework in the process. AI has made it easier to take on the challenge of writing apps in native code, according to the company. Quote of the week "You always want to go into a company that has the experience of having a significant cyber event in the past, because you don't have to argue what the risk is." New SolarWinds CISO Justin Henkel, articulating a pretty good career strategy for security executives and alluding to the massive supply-chain attack on its infrastructure that ensnared several government agencies six years ago. We're also reading: Inside Trump’s Aggressive Push to Build Data Centers on Public Lands : As reported by The Washington Sun: “I’m not against development or progress but I think there are too many questions about this and nobody is asking them,” said Mary Jo Rugwell, a former BLM state director who now serves as president of the Public Lands Foundation. “I think we’re just bending knee to the tech oligarchs and letting them do whatever they need to do.” Salesforce to Drop ‘Agentforce’ Branding From Some Products : The Information reports that next week's Dreamforce conference will be less Agentforce-y than last year's, which probably means Salesforce is also no longer planning to rename the company Agentforce https://www.businessinsider.com/salesforce-marc-benioff-says-he-might-rename-company-agentforce-2025-12?ref=thestack.technology . Mark Zuckerberg has regrets, obviously. Thanks for reading — see you Tuesday