I hit send and stare at my signature. The tagline. The bold claim. Two years ago I was a grad student who couldn't get a paper published. Now I'm signing emails with marketing slogans and venture capitalists are flying across the country to see my "office."
I don't know whether to laugh or panic.
Probably both.
The AC unit in the corner of the conference room squeaks to life, and the energy in the room shifts. Elad's associate - a guy named Amit who hasn't said much beyond nodding - leans forward and opens his laptop. The screen glows blue against his face as he pulls up a document I immediately wish I hadn't seen.
"We've been tracking this for six months," Amit says, his voice flat. "OpenAI and Anthropic report hundreds of daily interactions where users attempt to generate bioweapons. Reverse engineering nuclear technology. Step-by-step instructions for things that shouldn't be possible."
He clicks to the next slide. I see chat logs, redacted but terrifying. Prompt engineering attempts that bypassed safety filters. Users asking for "educational purposes" then pushing deeper, deeper, until the model spit out something genuinely dangerous.
"These models are capable of more than the public knows," Amit continues. "The companies downplay it. Internal metrics they don't publish. But we have sources."
I look at Jian. He's gone pale. Maddy's gripping her pen so tight her knuckles are white.