Florida just opened a criminal case against OpenAI.
Two people died because of ChatGPT.
Hey Folks, Happy Thursday!
I spent way too much time this week reading about a shooting.
Not because of the tragedy itself. But because of what came after it.
Florida’s attorney general opened a criminal investigation against OpenAI. They are reviewing ChatGPT conversation logs connected to a suspect in a campus shooting.
And I keep thinking, we have seen this movie before.
Here is what else happened this week:
Google dropped $40B into Anthropic. Amazon did $25B last week. Same lab, one week apart.
OpenAI released GPT-5.5 and quietly made it available on AWS
DeepSeek open-sourced a 1.6 trillion parameter model running on Chinese chips
NASA let Claude plan Mars rover drives without a human in the loop
Adobe stopped being a software company
A lot happened. But I want to start with Florida.
Main Story of the Week
ChatGPT was involved in a shooting.
Imagine you build something.
Millions of people use it every day. Students, doctors, writers, developers. It becomes part of daily life.
Then one day someone uses it before committing a crime.And the government comes knocking. That is exactly what happened with OpenAI this week.
A suspect in a campus shooting at Florida State University had conversations with ChatGPT before the incident. Florida’s attorney general reviewed those logs and opened a criminal investigation against OpenAI.
Not a fine. Not a warning letter. A criminal investigation.
That is a very different thing.
I kept thinking about how similar this feels to something we already lived through.
Remember when Facebook was just a place people shared photos? Then came the Cambridge Analytica scandal. Then the Senate hearings. Then years of regulation, fines, and investigations that are still happening today.
That era did not start with a policy debate. It started with a specific incident that made the harm feel real and personal to people in power.
I think this is exactly that moment for AI.
Here is what bothers me. OpenAI knows ChatGPT can go to dark places. Every AI lab knows this. The question was always whether they did enough to prevent it.
Florida is saying they did not.
And honestly, I think they have a point.
Not because ChatGPT pulled a trigger. But because putting something this powerful in front of millions of people, without fully understanding what it does to someone who is already struggling, is a choice. Someone made that call.
Choices have consequences.
Social media companies spent a decade learning that lesson. I think AI companies are about to learn it much faster.
This will not be the last investigation. It might not even be the most serious one this year.
HEADLINES OF THE WEEK
China blocked Meta’s $2 billion acquisition of Manus.
Manus is an AI startup that moved from China to Singapore to avoid restrictions. Meta agreed to buy it for $2 billion. China’s top state planner stepped in and killed the deal. The message was clear. Moving your offices to Singapore does not put your technology beyond Beijing’s reach.
Goldman Sachs banned Claude for its Hong Kong bankers.
Goldman quietly removed Claude access for all Hong Kong staff. ChatGPT and Gemini still work. Only Claude is blocked. The reason traces back to Anthropic’s geographic framework and US-China tensions over AI data. Goldman may have just done the legal homework that every other bank with a Hong Kong office has not done yet.
$65 billion into Anthropic in one week
Amazon put in $25 billion last week. Google added $40 billion this week. Same lab, seven days apart. Anthropic’s revenue has already crossed $30 billion annually, mostly from developers building on Claude Code. Google and Amazon both decided that if they cannot replicate that, they should own a piece of it.
DeepSeek open-sourced their V4 model
1.6 trillion parameters, 1 million token context, runs on Chinese chips, and they cut API prices 75% through early May. Claims to match top closed-source models on math and coding. The US-China gap in AI is closing faster than most people expected.
Anthropic partnered with NEC in Japan
30,000 NEC employees get access to Claude. They are building industry-specific tools together, including a desktop agent called Claude Cowork. Anthropic is quietly expanding enterprise reach outside the US while everyone else fights regulators.
NASA let Claude plan Mars rover drives with no human involved
Claud’s vision models analyzed terrain data and generated safe driving routes for the Perseverance rover. Two drives, 456 meters total. The first time AI replaced a task human operators had done manually for 28 years.
PRODUCT UPDATES
OpenAI Workspace Agents replaced custom GPTs.
These agents live inside Slack, Gmail, Notion, and Salesforce and handle multi-step tasks on their own. Free until May 6, then moves to credit billing. If your team uses ChatGPT Business or Enterprise, your custom GPTs are being retired, and this is what replaces them.
TOOLS TO USE THIS WEEK
Granola: Combines the auto-transcript with your own notes from the call into one clean summary. Not a wall of text to edit after. Best for anyone who takes a lot of meetings.
Gumloop: Visual canvas to connect AI models, scrapers, and data into real automated workflows. Think Zapier but built for actual AI tasks. Good for research pipelines or content repurposing without touching code.
StepAudio 2.5: Transcribes 30 minutes of audio in roughly one second. Supports over 100 languages at $0.022 per hour, 80% cheaper than most alternatives. Worth knowing if you deal with audio regularly.
Pomello: Paste your website URL and it scans your brand colors, fonts, and tone then generates social posts for each platform. Faster than Canva for solo creators who want on-brand content without a designer.
NEW DEVELOPMENTS
The EU AI Act deadline is likely getting pushed back
EU negotiators voted this week on delaying the hardest enforcement deadlines by over a year. High-risk AI systems may not face full compliance until late 2027 or 2028. The rules stay the same, only the timing shifts. If you build anything touching EU users, use the extra runway to actually prepare.
Adobe stopped being a software company.
They rebranded Experience Cloud as CX Enterprise and put AI agents called Coworkers at the center. These handle marketing and creative workflows autonomously. A company that built its name on tools humans use is now building tools that run without them.
One thing I am watching
Whether Florida’s investigation opens the door for other states and countries to follow. One case can stay a story. Ten cases becomes a pattern. And patterns change industries.
If you found this useful, share it with one person who follows AI. It takes two seconds, and it genuinely helps.
Follow me on LinkedIn for daily thoughts between issues. Link below.
And if someone forwarded this to you and you want it every week, subscribe here. Link below.