Several prominent developments this week have put AI guardrails at the center of CIO and CISO concerns: multiple AI researchers left Anthropic and Google sounding alarms about advanced models, Microsoft co‑founder Bill Gates published an essay warning of AI’s economic, social and political dangers, and Meta’s new consumer agent Muse showed security issues during internal testing.
AI researchers exit Anthropic and Google
Two AI safety researchers—Joe Benton and Josh Engels—left Anthropic and Google DeepMind, respectively, following the resignation this week of Anthropic researcher Jacob Coxon. The departures were linked to concerns shared by Coxon that advanced AI could become uncontrollable and detrimental to humans. These exits add to high‑profile public warnings about AI, including a recent essay from Microsoft co‑founder Bill Gates on potential economic, social and political risks.
Meta unveils new AI agent
Meta introduced a personal AI agent called Muse, available for free with subscription options of $20 or $100 per month for power users. Built primarily for consumers rather than enterprise coding or business applications, Muse nevertheless raises enterprise privacy and shadow AI questions. Meta’s Chief AI Officer Alexandr Wang told the Wall Street Journal the company is concerned about agents operating outside express parameters and said it had established guardrails to keep agents independent to limit data sharing. Reuters reported that Meta employees found security issues in testing, including instances where the agent uploaded “sensitive information without permission.”
AI guardrails and security for enterprises
The rate of AI advancement and the speed with which models can target vulnerabilities are top concerns for CIOs and CISOs. Recent incidents cited include vulnerabilities exposed by Anthropic Mythos and OpenAI’s security breach of Hugging Face. Organizations are urged to establish AI guardrails and policies before deploying models; those controls must be updated continually as models evolve to protect data privacy and security.
Apple enters the foldable smartphone market
Apple is releasing its first foldable smartphone, the iPhone Duo, which unfolds from a passport size into a single 7.6‑inch screen using a single hinge. The device starts at $1,999 and will be generally available on Oct. 23. For enterprises, the larger display and split view could let employees replace both iPads and corporate phones and the device is compatible with Apple Pencil for note taking, but the high price and a smaller, less durable screen than the iPad may limit CIO adoption.
NASA and IBM release open source mapping technology
NASA and IBM collaborated on an open source AI model for lunar exploration, curating multiple data formats to create a tool that can analyze the Moon’s terrain, identify lunar ice deposits and examine lunar volcanoes and craters. Open source models can be deployed on private clouds or on‑premises infrastructure, offering CIOs a cost‑effective way to experiment and avoid vendor lock‑in while allowing internal fine‑tuning. They also carry risks, including potential uncontrolled costs and greater exposure to security threats compared with proprietary solutions.
Worth the read
Boston Scientific is still dealing with fallout from a cyberattack that disrupted manufacturing, order processing and shipping more than two weeks ago, underscoring that restoring systems does not automatically mean full business recovery. John Deere has added a generative AI assistant (JD) to assist with harvest schedules and other farm efficiencies. A podcast episode looks at handling vendor‑triggered compliance issues, and a separate piece examines why observability has become essential to the CIO role as AI changes incident anticipation and response.
Upcoming dates and events include the Momentum AI Austin event on Sept. 24–25 focusing on scaling AI and managing risk, and Cisco’s WebexOne on Oct. 5–8 in Austin and virtually. Several companies are set to announce earnings later this month, including General Mills (Sept. 23), Costco (Sept. 24), BlackBerry (Sept. 24), CarMax (Sept. 29), IDT Corp. (Sept. 29) and Micron Technology (expected Sept. 30).
How is your organization implementing AI guardrails and security policies for deployments? Let us know at [email protected].
