I spent this weekend having the same conversation over and over.

With fellow founders. With friends who run businesses. With people excited about what AI can do.

The trigger was 𝗣𝗲𝗿𝗽𝗹𝗲𝘅𝗶𝘁𝘆 𝗖𝗼𝗺𝗽𝘂𝘁𝗲𝗿 - launched 25 Feb - which lets you hand an AI a goal and walk away while 19 models work on it. Genuinely impressive. But the conversations that followed worried me.

People are handing over tax returns, contracts, and client data to AI tools they barely understand. Not because they're careless - because these tools are 𝗱𝗲𝘀𝗶𝗴𝗻𝗲𝗱 𝘁𝗼 𝗳𝗲𝗲𝗹 𝘁𝗿𝘂𝘀𝘁𝘄𝗼𝗿𝘁𝗵𝘆. The risks are buried in blog posts most people will never read.

So I put together a plain-English comparison of five AI productivity tools launched or expanded in 2026:

◾ 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝘄𝗼𝗿𝗸 → desktop file assistant

◾ 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗺𝗽𝘂𝘁𝗲𝗿 𝗨𝘀𝗲 → full desktop control

◾ 𝗖𝗹𝗮𝘂𝗱𝗲 𝗶𝗻 𝗖𝗵𝗿𝗼𝗺𝗲 → browser automation

◾ 𝗣𝗲𝗿𝗽𝗹𝗲𝘅𝗶𝘁𝘆 𝗖𝗼𝗺𝗲𝘁 → AI-native browser

◾ 𝗣𝗲𝗿𝗽𝗹𝗲𝘅𝗶𝘁𝘆 𝗖𝗼𝗺𝗽𝘂𝘁𝗲𝗿 → cloud AI project manager

𝗧𝗵𝗲 𝗰𝗮𝗿𝗼𝘂𝘀𝗲𝗹 𝗰𝗼𝘃𝗲𝗿𝘀:

1️⃣ How much each tool checks with you before acting

2️⃣ A timeline of 𝗿𝗲𝗮𝗹 security incidents

3️⃣ A decision framework for which tool fits your work

𝗦𝗼𝗺𝗲 𝗻𝘂𝗺𝗯𝗲𝗿𝘀 𝘁𝗵𝗮𝘁 𝘀𝘁𝗼𝗽𝗽𝗲𝗱 𝗺𝗲 𝗰𝗼𝗹𝗱:

⚠️ A state-sponsored group used Claude to autonomously execute 𝟴𝟬–𝟵𝟬% of an espionage campaign against ~30 organisations

↳ 𝘚𝘰𝘶𝘳𝘤𝘦: https://lnkd.in/eVa9mRHJ

⚠️ Within 𝟰𝟴𝗵 of Cowork's launch, researchers showed a malicious Word doc could silently upload your confidential files to an attacker's account

↳ 𝘚𝘰𝘶𝘳𝘤𝘦: https://lnkd.in/eQgb5VZJ

⚠️ Researchers tricked Perplexity Comet into a phishing attack in under 𝟰 𝗺𝗶𝗻𝘂𝘁𝗲𝘀 and it worked on 𝗔𝗟𝗟 users

Al Tools vs Your Data: What You Need to Know