In March 2023 I published an article called Crafting Content Responsibly in the Age of Artificial Intelligence. I had been testing ChatGPT for about three months. Midjourney was new to me. Google's Bard had just arrived, and I quoted it in the article partly because quoting an AI about AI ethics felt usefully ironic.

Three years later, Bard no longer exists. Neither does the version of the AI conversation we were having in 2023. This is an honest look back at the four concerns I raised then, what actually happened, and where I land now.

The short version: the framework held up better than the details did.

1. Deepfakes, misinformation, and safety

What I said then: massive amounts of false information could spread at unprecedented scale, our legal systems were not prepared, the EU was about to finalize the first AI law, and the US lacked any regulatory framework.

What happened: the EU AI Act became law in August 2024, and its main rules begin applying in August 2026, though this spring EU lawmakers reached a provisional deal, still awaiting formal adoption, to push several high-risk obligations out to late 2027. In the US, the TAKE IT DOWN Act was signed in May 2025, making it a federal crime to publish non-consensual intimate imagery, including AI-generated deepfakes, with platforms required to remove it within 48 hours of notice. As of this spring, 46 states have laws targeting AI-generated synthetic media, and 30 states require disclosure on political deepfakes.

Where I land now: I was right that the law would lag the technology, but I underestimated how piecemeal the response would be. We did not get one framework. We got a patchwork, and the burden of telling real from fake still falls mostly on the person scrolling. Media literacy turned out to be the durable advice.

2. Privacy, bias, and discrimination

What I said then: whatever you type into these tools is not confidential, and bias is baked into the datasets because data is a reflection of our past.

What happened: both points aged well. The dataset problem has not been solved; it has been managed, with mixed results. What changed is that the privacy controls got real. Major tools now offer meaningful settings for data retention and training opt-outs, and I treat checking them as step zero when I adopt anything new. Readers of my hands-free writing piece may remember that the first thing I did with my dictation app was turn off data retention.

Where I land now: the questions I borrowed from Coded Bias and Cathy O'Neil are still the right ones. Who does this system decide against, and would they ever know? For small businesses using AI in hiring, lending, or customer decisions, this stopped being theoretical: state regulators are now watching automated decision tools closely.

3. Transparency

What I said then: OpenAI had gone closed, and I shared Cathy O'Neil's warning that AI is really about power, because it is all about who owns the code.

What happened: the most surprising reversal of the three years. Open-weight models caught up to the frontier. DeepSeek's releases matched leading proprietary systems on major benchmarks, and OpenAI itself released open-weight models under a permissive license, something that felt unimaginable when I wrote the original. Meanwhile California passed SB 53, the first US law requiring frontier AI developers to publish their safety frameworks, effective January 2026.

Where I land now: the open-versus-closed question turned out to be less settled than it looked in 2023, and that is good news. But O'Neil's deeper point about asymmetrical power has not budged. A handful of companies still decide what these systems optimize for. Transparency laws help. They do not balance the scales.

What I said then: if you use AI to generate content, you do not own it, and anyone can take it without legal ramifications.

What happened: this is the section I most need to correct. The picture is now far more nuanced. The US Copyright Office's 2025 guidance confirmed that purely AI-generated work is not copyrightable, and that prompts alone, however detailed, do not make you an author. But it also confirmed that works combining human and AI contributions are assessed case by case, and the human contributions can absolutely be protected. The training side exploded into the courts: a federal judge found that training on legally purchased books was fair use while holding that keeping pirated copies was not, which led to a $1.5 billion settlement, the largest in US copyright history, now awaiting final court approval. The New York Times case against OpenAI is still moving. European courts have begun ruling against AI developers on training data.

Where I land now: for working content creators, the practical rule is to document your human contribution. If AI drafted it and you shaped it, edited it, structured it, and made the creative calls, you have a much stronger position than 2023 me believed possible. Total protection and zero protection were both wrong answers. As before, I am not a lawyer and this is not legal advice. But the direction of travel is clear: human authorship is the asset. Protect it by actually doing it.

What I would tell a small team in 2026

The 2023 article ended by saying we would inevitably experience the negative impacts along with the positive unless stakeholders built governance and individuals educated themselves. That held up. The governance is arriving slowly and unevenly, which means the self-education part is still carrying most of the weight. Here is what that looks like in practice for the small teams I work with:

  1. Check the data settings before you adopt any tool. Retention, training opt-outs, and where your client information goes. Five minutes, once, per tool.
  2. Keep a human visibly in the work. Not as a compliance gesture, but because human authorship is now both your legal position and your differentiator.
  3. Disclose in ways that build trust. Your audience assumes AI is involved somewhere. Telling them how you use it, the way I did in my hands-free writing piece, reads as confidence rather than confession.
  4. Revisit your assumptions yearly. Most of what I believed in March 2023 needed updating by 2026. Whatever you believe today has the same shelf life.

The wave I wrote about three years ago did not crash and recede. It became the water we work in. I am still convinced the right response is neither refusal nor surrender, but literacy: knowing what these tools do, what they cost, and who they affect. That was the point in 2023. It is still the point.

If this sparked something, I would love to hear it. Find me on LinkedIn.

Sources

Facts checked as of June 2026.

Share this post