One-Line Summary

OpenAI makes web-informed image generation available in ChatGPT while policy scrutiny and mega-infrastructure bets (and hiring raids) signal a harder, more regulated road for everyday AI work.

New Tools

OpenAI’s updated image generator can now pull information from the web

ChatGPT’s built-in image generator gets a major upgrade: with Images 2.0 and a selected Thinking model, it can consult the web to help craft a consistent set of visuals from one prompt. OpenAI also improves instruction-following and text rendering in images, aiming to make creative work faster and more controllable. ¹

When a Thinking model is used, Images 2.0 can generate up to eight images at once, keep characters/objects/styles consistent across scenes, and create visual explainers from uploaded files. It also adds more aspect ratios (from 3:1 to 1:3), higher resolution up to 2K, and better support for text in Japanese, Korean, Chinese, Hindi, and Bengali. ²

Access and limits matter for teams: paid tiers can pick between GPT‑5.3 Instant and GPT‑5.4 Thinking, with manual Thinking usage capped at up to 3,000 messages per week on Plus and Business. Message caps include Free at 10 every 5 hours (GPT‑5.3), and Plus at 160 every 3 hours. Context windows vary: Instant is 16K (Free), 32K (Plus/Business), 128K (Pro/Enterprise); Thinking offers 256K on paid tiers and 400K on Pro, plus a new thinking‑time toggle to balance speed vs depth. ³

The launch lands as rivals push research agents: Google’s new Deep Research and Deep Research Max (built on Gemini 3.1 Pro) automate web‑sourced, cited analysis and support the Model Context Protocol, showing how fast “agentic” tasks are evolving around search and synthesis. ⁴

Big Tech

Anthropic outspends OpenAI in biggest-ever lobbying quarter

This item is about what two AI labs spent in Washington in Q1 2026: Axios reports Anthropic spends $1.6 million and OpenAI $1 million, both hitting their biggest‑ever lobbying quarters. That puts dedicated AI labs closer to Big Tech’s long‑standing policy footprint. ⁵

Meta remains far higher at $7.1 million for the quarter, with Amazon at $4.4 million and Google at $2.9 million. Axios notes Anthropic’s topics include AI procurement, Defense Department procurement, supply chain risk, and acceptable‑use policy; OpenAI’s filing highlights AI and copyright, cybersecurity, cloud, and infrastructure. ⁵

Behind the policy push is massive scale: Anthropic says it commits more than $100 billion over ten years to AWS, securing up to 5 gigawatts of capacity (including Trainium 2/3) to train and deploy Claude, now used by over 100,000 AWS Bedrock customers; Anthropic also says run‑rate revenue surpasses $30 billion. ⁶

Florida to open criminal investigation into OpenAI over ChatGPT’s influence on alleged mass shooter

Florida’s attorney general, James Uthmeier, opens a criminal investigation into OpenAI, saying the probe examines whether ChatGPT “offered significant advice” to an alleged FSU campus shooter; subpoenas seek internal policies and training materials on threats of harm and law‑enforcement cooperation. ⁷

A Yahoo report (citing NBC News) includes OpenAI’s response that last year’s FSU shooting was a tragedy but that ChatGPT is not responsible; the company says the chatbot provided factual information available across public sources and did not encourage illegal or harmful activity. ⁸

The inquiry tests emerging liability questions for AI outputs, with investigators referencing specific pre‑incident queries from the suspect; legal theories and timelines are not yet detailed, underscoring unsettled ground for chatbot accountability. ⁹

Meta hires five Thinking Machines Lab founders including a reported $1.5 billion engineer

Meta is reported to have hired five founding members of Mira Murati’s Thinking Machines Lab; The Next Web says co‑founder Andrew Tulloch joins with a compensation package reportedly worth $1.5 billion over six years, highlighting the escalating costs of frontier AI talent. ¹⁰

Alongside hiring, Meta expands infrastructure, breaking ground on an AI‑optimized data center in Tulsa, Oklahoma, an investment of over $1 billion that will support around 100 operational jobs when complete and add clean‑energy‑matched capacity. ¹¹

Separately, iPhone in Canada (citing Reuters) reports Meta installs software on U.S. staff machines to capture mouse movements, clicks, and keystrokes within a defined set of work apps to train “computer‑use” AI agents, with stated safeguards and no use in performance reviews. ¹²

What This Means for You

OpenAI’s Images 2.0 is built for production tasks like multi‑panel social posts, storyboards, or manga pages: generating up to eight consistent frames per prompt and sharper text lets marketers, designers, and content teams draft campaign sets in one pass instead of stitching styles by hand. That can shorten rounds of review and keep brand elements aligned across formats. ²

Usage tiers and toggles matter operationally. If your team needs reasoning‑heavy image sets, plan around the Thinking message cap (up to 3,000/week on Plus/Business) and pick a thinking‑time setting (fast vs. deeper) before deadlines. Context windows also drive feasibility for long briefs: 256K on paid tiers and 400K on Pro (Thinking) can hold large creative specs without constant re‑prompting. ³

The Florida case signals rising legal scrutiny on how teams use chatbots around safety‑related topics. Keep clear guidelines for sensitive prompts (weapons, self‑harm, criminal activity), log who uses AI for what, and document your safety policies—so you can show due diligence if questions arise. ⁷

On strategy, Anthropic’s $100B AWS commitment and Washington spending suggest AI platform choices are increasingly about capacity, policy, and reliability, not just features. When you pick a vendor, weigh access to compute and enterprise controls alongside model quality; these factors shape uptime, quotas, and support. ⁶

Action Items

1. Try ChatGPT Images 2.0 for a series post: Give one prompt and generate up to eight images with a consistent character/style; test both wide (3:1) and tall (1:3) ratios for your next social carousel or storyboard.
2. Use the Thinking toggle on a hard task: In ChatGPT, switch to GPT‑5.4 Thinking and set Extended thinking‑time for a complex brief (e.g., a product explainer with visuals), then compare output vs. Standard.
3. Set internal AI safety guardrails: Write a one‑page guideline for staff covering prohibited or sensitive prompts (weapons, self‑harm), escalation paths, and how to capture AI‑assisted work in project logs.
4. Plan around message limits: Map your team’s busiest windows against ChatGPT caps (e.g., Plus 160 messages/3 hours; Thinking manual cap) and decide who gets Pro/Business seats for heavy weeks.
5. Benchmark multilingual text rendering: If you publish in Korean or Japanese, test Images 2.0 text fidelity by generating ad mockups or thumbnails that include short headlines in target languages.