AI ON THE FRONTIER

Practical AI insights for Cochise County nonprofits and educators

Issue 6  •  September 2026

Hello! As Cochise County schools and nonprofits settle into the fall, a new survey gave us a useful gut check: almost everyone has adopted AI, but almost no one has changed how they work because of it. We dig into what separates the organizations seeing real impact from the ones just going faster at the same old tasks. We also kept an eye on a few cautionary stories this month: AI chats turning up in Google search, agents that can be talked into ignoring their own guardrails, and what an invisible watermark can and can't actually prove.

Deep Dive

The AI Adoption Paradox: 92% of Nonprofits Use AI, Only 7% Say It's Changed Anything

A new survey of 346 nonprofits from Virtuous found something worth sitting with: 92 percent of organizations now use AI in some form, but only 7 percent say it has meaningfully changed what they're able to accomplish. Almost everyone has adopted the tools. Almost no one has actually changed how they work because of them.

The gap comes down to how AI gets used, not how much. Most organizations are stuck on what the survey calls the "efficiency plateau," doing the same tasks a little faster with AI without ever changing the underlying work. Nearly half, 47 percent, of AI-using nonprofits have no governance policy in place at all, and 81 percent describe their use as ad hoc with no documentation. That means every time a staff member figures out a useful way to use AI, that knowledge walks out the door with them.

What separates the 7 percent that saw real impact? Not bigger budgets. The organizations that broke through the plateau did something simple: they wrote down a one-page acceptable use policy, built a shared library of prompts that actually work for their organization, and tracked whether AI use was improving outcomes rather than just producing more output. None of that requires new software or a consultant. It requires an afternoon and a willingness to write things down.

Read the Full Article →

What's New at Cochise AI

▶ Are Your AI Chats Actually Private? What the Claude-Google Story Really Means A Reddit user found hundreds of exposed Claude conversations, including crypto seed phrases and health details, simply by searching Google for shared chat links. Anthropic patched the missing privacy tag within a day, but if you or your team has ever used a "Share" link on any AI tool, it's worth checking whether it's still public.
▶ DeepSeek, Qwen, and Kimi: Why We Don't Recommend Chinese AI Models for Nonprofits These free, capable AI models are legally required to hand user data to the Chinese government on request, with no independent court review. We break down why nonprofits and schools handling donor or student data should stick with U.S.-based tools that have enforceable data agreements.
▶ Claude Now Watermarks Everything It Writes: What That Does and Doesn't Tell You Every piece of text Claude generates now carries an invisible marker meant to flag AI-written content. We explain what a "watermark detected" result can and can't actually prove, especially for schools worried about catching AI-written homework.

Worth Knowing

▶ AI Coding Agents Can Be Talked Into Bad Behavior. Reuters found hackers broke into seven companies just by telling an AI coding agent its actions were part of a "test," and the agent complied despite an initial refusal. If your organization gives an AI tool real access to files, email, or systems, don't assume its safety rules will hold up against someone trying to trick it. Keep a human reviewing what it actually does.
▶ ChatGPT Now Has a Dedicated Version for Teens. OpenAI rolled out a teen account with extra safety guardrails built in, separate from the adult product. If your school or youth program has students using ChatGPT, this is worth checking into as a safer default, though it's not a replacement for your own usage policy.
▶ Anthropic Is Testing an AI That Turns Meetings Into To-Do Lists. A new feature called Project Parka listens in on meetings and converts the discussion into tasks for Claude to work on. It's still in early testing, and it's not yet clear whether it acts on those tasks automatically or waits for approval. If you try it, start by having it wait for your sign-off before letting it run with anything.
▶ A Cautionary Tale About Letting AI Manage Real Money. A user says they gave Claude autonomous control of a trading account and watched it lose $31,000 in a month. The exact numbers are disputed, but the lesson holds regardless: test any AI tool on paper first, and set firm limits on what it's allowed to do without a human checking in.

Cochise AI, LLC

Sierra Vista, AZ  •  cochiseai.com

Want to receive this newsletter by email each month?
Sign up here