OpenAI Locks Down Agents Against Data Theft

OpenAI just shipped Lockdown Mode for ChatGPT — a security setting that blocks a class of attack that has quietly drained data from AI deployments for over a year.

Why it matters: Prompt injection attacks, where malicious content in a document or webpage tricks an AI agent into exfiltrating sensitive data, have been the dirty open secret of enterprise AI rollouts. Lockdown Mode disables live web browsing, agent mode, deep research, image retrieval, and file downloads in one switch. It's now live for all logged-in ChatGPT users across every plan tier.

The GTM angle: If your revenue team is using ChatGPT agents to process inbound emails, summarize call recordings, or research accounts, your IT or security team now has a sanctioned way to harden those workflows. Get them to enable it for any deployment touching customer data.

• Lockdown Mode trades capability for safety: agents can't browse or download files when it's on. Teams need to decide per-workflow whether that tradeoff makes sense.
• This is OpenAI responding directly to documented exploits — not a proactive measure. The threat it addresses is real and already in the wild.

Go deeper: https://thenextweb.com/news/chatgpt-lockdown-mode-prompt-injection

Trump's NSPM-11 Gives the Military a Kill-Switch Override on AI Vendors

President Trump signed a directive Friday ordering military and intelligence agencies to fast-track advanced AI adoption — and banning any AI vendor from pulling, degrading, or modifying a deployed system without government approval.

Why it matters: National Security Presidential Memorandum 11 (NSPM-11) creates a "rapid onboarding" framework for AI models across the Defense Department, with a hard rule: once the government deploys your AI, you cannot disable or alter it unilaterally. That's a fundamental shift in the vendor-customer relationship for any company selling into federal markets.

The GTM angle: If your company is targeting government or defense accounts, or competes with vendors who do, this is the policy environment you're selling into. Compliance, explainability, and continuity guarantees are now table stakes for federal AI deals — not differentiators.

• France is simultaneously field-testing Arcadia, its own European AI command system, as a direct alternative to Palantir's Maven — a sign that the AI defense market is splintering along national lines.
• Any AI vendor with federal ambitions should be reviewing contract terms around system availability and modification rights now.

Go deeper: https://thenextweb.com/news/trump-ai-military-memo-autonomous-weapons

Google Is Paying SpaceX $920 Million a Month for 110,000 GPUs

Google signed a deal to rent 110,000 Nvidia GPUs from SpaceX starting October 2026, at $920 million per month — the second major SpaceX data center deal in weeks, timed ahead of SpaceX's June 12 IPO.

Why it matters: The GPU supply crunch is real enough that Google, which builds its own AI chips, is paying nearly $11 billion a year to lock in Nvidia compute through an aerospace company that's rapidly becoming a cloud provider. That's what tight supply looks like when the biggest players are stacking capacity wherever they can find it.

The GTM angle: For revenue teams evaluating AI vendors, this is a useful data point: the cost of AI infrastructure is not coming down soon, and any vendor pricing that feels aggressive today is likely being subsidized. Assume AI API costs stay elevated through at least mid-2027 when modeling out build-vs-buy decisions.

• SpaceX's projected annual data center revenue is set to exceed its combined revenue from Starlink, launch services, and AI in 2025 — a remarkable shift for a rocket company.
• Google's willingness to pay a premium through a third party signals that even hyperscalers can't secure enough Nvidia supply through normal channels.

Go deeper: https://www.tomshardware.com/tech-industry/artificial-intelligence/google-signs-usd920m-monthly-compute-deal-with-spacex-companys-projected-annual-data-center-revenue-to-exceed-its-combined-proceeds-from-starlink-launch-services-and-ai-in-2025

OpenAI Is Rebuilding ChatGPT Around Agents, Not Chat

A senior OpenAI employee said "chat is dead," and the company is weeks away from shipping a major ChatGPT overhaul that reframes it as an agent platform, not a conversation interface.

Why it matters: The Financial Times reports the redesign is imminent. OpenAI has been publicly positioning ChatGPT as something that does work, not just answers questions. If this ships the way insiders describe, the interface your teams use every day will look and behave fundamentally differently — built for multi-step tasks and autonomous workflows, not back-and-forth Q&A.

The GTM angle: Revenue teams that have built ChatGPT habits around prompt-response workflows should start piloting agent-mode tasks now, before the redesign makes the old interface awkward. The teams that are ahead on agentic workflows won't need to retrain in September.

• OpenAI is also still building a "super app" — a broader consumer product that extends well beyond AI chat. The ChatGPT overhaul is a step in that direction.
• This is a platform-level shift, not a feature update. Your enablement materials for ChatGPT are going to need a rewrite.

Go deeper: https://techcrunch.com/2026/06/07/openai-is-still-working-on-that-super-app/