In the last 11 days, we tracked 5,239 changes across 10,304 AI tools: new AI features, model and subprocessor changes, new data retention terms, security incidents, and much more. Is the UN really doing enough to protect us from AI risks?
No.
Things aren’t slowing down. So the work is making it safe instead. That’s what we do.
We have been watching this since 2023, when we started as LLM threat researchers finding novel vulnerabilities in everyday tools like Claude, Copilot and Slack AI. We’ve been cited by OpenAI, Forbes, The Information, and much more. One thing we’ve noticed: the risks of AI are increasing, and at a rapid pace.
We surface new threats every single day to protect you and your companies. Here are just 5 examples.
1. Labcorp
6/29/2026
You know, the company in charge of your private health data? On June 29, 2026, we found this added to their privacy notice:
“Labcorp may collect, use and sell Personal Data for training of AI to the extent permitted by applicable law.”
Talk about dystopian.
2. Gamma
9/17/2026
Its FAQ used to say, word for word:
“No, Gamma employees do not access your decks without your explicit permission.”
On September 17, 2026, we found that sentence deleted. Now employees can access without your permission.
Sneaky.
3. Lovable
8/6/2026
On August 6, 2026, we found that starting September 9, Free and Pro customer data may be used for AI model training.
Not very loving.
4. Atlassian
4/16/2026
April 16, 2026, we found Atlassian using customer metadata and in-app data to train models and improve its own services.
Opted in by default, and retained for up to seven years (seven!).
5. Zoom
8/5/2026
On August 5, 2026, we found that Zoom’s AI could be hijacked through a malicious add-on and made to pull meeting transcripts, messages and data from connected apps.
Seriously, not cool. We sent our customers the fix that day.
What to look out for
These are the four changes we see most often.
Data training. Your data gets used to teach an AI model, often switched on by default.
New AI in old tools. A tool you approved years ago adds AI without you knowing.
Nth-party risk. Your vendor’s vendor has a new risk, and it becomes your problem indirectly.
Security vulnerabilities. A flaw in the tool that puts you at risk.
What we do
Your tools’ AI risks continue to change. We help you bring new tools on safely, and watch the ones you already have. Before you adopt a tool, we tell you what AI is in it, what is at risk, and what a standard vendor review would miss. Once you are using it, we alert you to changes and give configuration and policy recommendations to mitigate the risk. Harvey’s change was reversed because people knew about it in December, and not in a contract review two years later.