Mitigating the Risks of AI

1. Mitigating the Risks of AI

Welcome back! Now that we've identified the key risks of working with AI, it’s time to explore practical strategies for using it safely and responsibly. Think of these as your AI safety protocols—simple habits that protect both you and your organization.

2. Mitigating Knowledge Fabrication

Let’s start with knowledge fabrication—those convincing but false facts that AI can produce. The key strategy is a “trust but verify” mindset. Always double-check critical information, especially in four areas: numbers and data points, citations and references, historical facts and dates, and technical details. If it claims that “73% of consumers prefer your product,” don’t take it at face value—track down the study. If it cites a paper or article, confirm that it exists and says what’s claimed. You’re not distrusting AI; you’re treating it like a colleague whose work you review before presenting it. A quick verification can prevent costly or embarrassing errors.

3. Mitigating Recency Ignorance

Next, recency ignorance. Remember that AI lacks awareness of information released after its knowledge cutoff date. Always check what that cutoff is; most models will tell you if asked. Then, verify anything that happened afterward, especially prices, regulations, personnel changes, and news. For instance, if your model’s knowledge stops in April 2025, and you need information on a September 2025 product launch, supplement it with web searches or other current sources. Some AI tools now have built-in web browsing—use that when its available.

4. Mitigating Biased Outputs

Now let’s talk about biased outputs. AI models reflect the biases in their training data, so you need to review outputs critically. Focus on three areas: hiring materials, marketing content, and decision-making tools. After generating text, ask yourself, “Could this discourage or exclude anyone?” For example, a job post emphasizing “aggressive” communication might unintentionally deter strong candidates who value collaboration. You can also ask, “Could this be biased?” and have it revise. When possible, invite colleagues to review sensitive content. Many companies also use bias-detection tools—use them if available.

5. Mitigating Sycophantic Outputs

Now onto sycophantic outputs. Mitigating sycophancy requires you to actively guide the AI toward objectivity and critical analysis. Maintain active skepticism, especially when its response feels particularly validating. When you need critical feedback, explicitly ask for it: “What are the potential problems with this approach?” Or “What are some alternative perspectives on this issue?" The practical wisdom here is: if the response seems too convenient or agreeable, pause and consider whether you might benefit from a more critical perspective.

6. Mitigating Privacy and Data Exposure

Privacy and data exposure require the most vigilance. Never share four categories of sensitive information: passwords or credentials, customer personal data, particularly PII, proprietary company information, or confidential strategies or plans Before submitting any prompt, pause and ask: “Should I remove any sensitive data?” When necessary, anonymize—replace real names with placeholders, strip identifying details, and use generic examples. Check your organization’s policies and, when possible, use enterprise-grade AI tools that don’t train on your data. Here’s a simple rule of thumb: if you wouldn’t post it publicly, don’t paste it into a standard AI tool.

7. Your AI Safety Checklist

To bring it all together, use this AI safety checklist before submitting any prompt:

8. Your AI Safety Checklist

Am I sharing sensitive information?

9. Your AI Safety Checklist

Can I verify factual outputs?

10. Your AI Safety Checklist

Could the result be biased?

11. Your AI Safety Checklist

And is the information time-sensitive? These questions take seconds to consider but protect against the biggest risks. Make this checklist a habit—

12. Your AI Safety Checklist

like looking both ways before crossing the road. With practice, it becomes automatic. These safety protocols aren’t meant to slow you down, but to help you use AI confidently and effectively while safeguarding yourself and your organization.

13. Let's practice!

Now it’s your turn to put these strategies into practice!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.