OpenAI rolls out safety kill switch, AI agents, and shopping tools

Control Room: OpenAI’s 35-second safety kill switch

OpenAI’s Control Room can shut down an AI model in 35 seconds if it detects safety or security risks. The tool monitors every workload—training runs, evaluations, data processing—across OpenAI’s research infrastructure. If something looks off, it blocks the output before it reaches users.

Control Room ties together three components: the workload (jobs like reinforcement learning experiments), the inference engine (which executes the job), and a central control service. Engines must check with the control service before fulfilling any request. If the job is flagged, the service denies it. OpenAI also runs continuous scans to catch rogue engines that might try to bypass the system.

Business Hub and AI agents take on Google’s search dominance

OpenAI opened Business Hub in beta, letting companies create verified profiles on ChatGPT. Businesses can update product details, prices, and stock, while tracking aggregated insights like profile views and website clicks. Verification is required, but OpenAI hasn’t revealed how it works.

Business Agents, available to US e-commerce and online-service firms, act as AI-powered reps for brands. These agents interact with ChatGPT’s 1.2 billion weekly users, answer questions, and pull real-time product info. They can also be sponsored as ads—OpenAI’s ad revenue already exceeds a $1 billion annual run rate.

Every Business Agent must disclose it’s AI, ensure price accuracy, and avoid pressure tactics. Users must explicitly approve any purchase or data sharing.

OpenAI retires older voice modes amid user backlash

OpenAI will retire Advanced and Standard voice modes on December 11. All users will switch to GPT-Live, introduced in July, which allows real-time interruptions and has nearly doubled voice usage since launch.

GPT-Live listens and speaks at the same time, speeding up conversations. It even inserts filler words like ‘mhmm’ to mimic natural dialogue. But many users hate it. Complaints on Reddit and X highlight frustration over interruptions, tone shifts, and nostalgia for older voices like Vale.

This isn’t the first time OpenAI has tried to phase out older voices. In September 2025, it reversed a plan to retire Standard Voice after similar backlash. For now, live video, screen-sharing, and custom GPT voice chats will still use Advanced mode until the December cutoff.

Daybreak: AI tools for hacking and patching

OpenAI unveiled two cybersecurity tools under the Daybreak brand. Daybreak Red Team uses AI agents to probe web apps for vulnerabilities, matching human penetration testers in speed and depth. Daybreak Defense Factory automates vulnerability scanning and generates patches. Both run on OpenAI’s GPT-6 Cyber model.

No pricing has been announced, but Daybreak Red Team is available in early access. Nancy Wang, OpenAI’s head of Cyber-Applied AI, says the tool brings ‘human-level’ precision to security testing.

ChatGPT Wallet: AI agents that spend your money

OpenAI launched ChatGPT Wallet, letting AI agents called dots shop and check out for users. The feature is live for a limited group of US-based Pro subscribers aged 18+ with a Visa, Mastercard, or American Express card.

Users must approve every purchase via Face ID, fingerprint, or passcode. OpenAI uses payment tokens to shield card details from merchants. Partners like Visa ensure users keep their rewards points and fraud protections.

For now, OpenAI isn’t taking a cut of transactions. But companies like Amazon, Delta, and United Airlines have banned AI agents from making purchases, citing security and regulatory risks.

Google and Amazon face new competition from AI agents

OpenAI’s Business Hub and Business Agents now compete directly with Google’s My Business, the default platform for consumer discovery. This follows OpenAI’s March retreat from an e-commerce hub that charged commissions—a model it has abandoned in favor of verification and sponsorships.

The move puts pressure on startups like Peec AI, which track brand visibility in AI chats. If OpenAI absorbs similar features, third-party tools could become redundant. It also raises the stakes for competitors like Google Cloud, which recently debuted autonomous AI agents with voice-powered guardrails.

Amazon’s ban on external AI agents highlights a growing divide. While OpenAI pushes agent-driven commerce, incumbents are tightening control—setting up a clash over who owns the future of AI-powered transactions.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

OpenAI’s Business Hub and Business Agents directly challenge Google’s dominance in consumer discovery by turning ChatGPT into a rival platform for e-commerce and local search. By avoiding transaction fees and focusing on verification and sponsorships, OpenAI avoids repeating its earlier misstep with a commission-based e-commerce hub. This move could also reduce demand for third-party tools that track brand visibility in AI chats, as OpenAI builds similar features into its core product.

Key takeaways

• Control Room sets a new standard for AI safety, potentially pressuring competitors to adopt faster intervention systems.
• Business Agents position ChatGPT as a discovery platform, eroding Google’s advantage in e-commerce and local search.
• Retiring older voice modes shows OpenAI prioritizing scalability over user nostalgia, despite backlash.
• ChatGPT Wallet normalizes AI-driven purchases, even as major retailers resist—setting the stage for regulatory and security debates.

Related on Lazyfounder

• OpenAI
• Google
• Amazon
• X
• Google Cloud’s AI agents now work autonomously with voice and guardrails
• United Airlines bans AI agents from booking flights over security and regulatory risks