AI moves fast. Stay in the know.

Google is rolling out Spark, an always-on AI agent inside the Gemini app that can run continuously in the background and take actions on a user’s behalf. Business Insider reported that Spark can read and use information from Google products such as Gmail, Docs, Sheets, and Slides, with future plans to connect to third-party tools. While the agent is positioned as a productivity assistant, the enterprise risk is significant: once AI agents can access workplace data and act across connected systems, businesses need stronger controls around permissions, monitoring, data exposure, and accountability.

Source: Business Insider

What to know:

• Google described Spark as an always-on AI agent that can run 24/7 and continue working even when a laptop is closed.
• The agent is powered by Gemini 3.5 and runs on Google Cloud, allowing it to operate continuously in the background.
• Spark can take actions on behalf of users, including planning tasks, collating notes, drafting emails, and managing reminders.
• Google said Spark will be able to read and use information from first-party products such as Gmail, Docs, Sheets, and Slides.
• The company also plans to connect Spark with Chrome and eventually third-party tools, expanding the agent’s access across workflows.
• For businesses, this creates new governance concerns around what data AI agents can access, what actions they can perform, and how those actions are monitored.

Why it matters:

As Gemini becomes embedded into everyday workplace systems, always-on AI agents introduce a new governance challenge by shifting from simple assistance to active task execution. Once an AI agent can read emails, scan documents, pull information from spreadsheets, draft messages, and act across connected apps, organizations need clear visibility into what the agent accessed, what it generated, and whether its behavior aligns with internal policy. Without AI observability and governance controls, businesses may face risks tied to sensitive data exposure, unauthorized actions, inaccurate outputs, shadow AI usage, and weak accountability when AI-driven tasks affect customers, employees, or business operations. 

Researchers and users are raising concerns after reports revealed that AI chatbots, including ChatGPT and Gemini, surfaced real phone numbers and sensitive contact details in generated responses. Security experts warn that attackers may also be manipulating online content to feed fake support numbers and fraudulent information into AI systems, creating new risks around AI trust, privacy, and misinformation.

Source: New York Post

What to know:

• Users reported cases where AI chatbots exposed real phone numbers and personal contact information in responses.
• Researchers warned that attackers are poisoning AI outputs with fake support numbers and malicious contact details.
• Retrieval-based AI systems can surface inaccurate, outdated, or sensitive information from across the web without proper verification.
• Unlike search engines, conversational AI tools often present responses with high confidence, making users more likely to trust incorrect information.
• Experts warn enterprises could face higher risks of phishing, fraud, social engineering, and unintended data exposure as employees increasingly rely on GenAI tools.
• The incidents are increasing calls for stronger AI governance, output monitoring, and enterprise safeguards around AI-generated responses.

Why it matters:

For mid-sized businesses adopting GenAI, the risk is not just misinformation but loss of visibility into what AI systems are surfacing and why. If AI tools can confidently expose manipulated or sensitive information, organizations face growing threats tied to data leakage, compliance risks, phishing, and reputational damage. Businesses will increasingly need AI observability, output validation, and governance controls to monitor how AI-generated information is being used across the organization.