Sam Altman Says We Should Accept 'Bad Things' From AI. I'm Not So Sure.

The short version

OpenAI’s CEO, Sam Altman, said we should be prepared to accept “some bad things” from AI in exchange for its massive benefits. This forces a difficult conversation about what level of risk is okay and who gets to decide. Frankly, it’s a statement that’s a lot easier to make when you’re the one building the technology, not the one potentially living with its consequences.

What did Sam Altman actually say?

I’ve been chewing on this all day. In an interview, Sam Altman made the case for a more aggressive, risk-tolerant approach to deploying artificial intelligence. He argued that the potential upside is so enormous that it justifies absorbing some negative impacts along the way. This wasn’t some off-the-cuff remark; it was a clear articulation of a philosophy. It’s a way of thinking that directly shapes the tools millions of us are now using every single day.

According to the Forbes report on his comments, Altman is trying to carve out a position distinct from competitors like Anthropic, a company known for its more cautious, safety-focused public stance. Where Anthropic talks about building constitutional safeguards into its models, Altman is advocating for a “lighter-touch” regulatory environment. He believes this will allow innovation to flourish. He also expressed skepticism about the most extreme, doomsday predictions for AI, suggesting the focus on existential risk is a distraction from getting powerful tools into the world now.

This isn’t a new idea. It’s the classic Silicon Valley mindset. Move fast and break things. The problem is that the ‘things’ at stake are no longer just servers or software bugs. They’re people’s livelihoods, the integrity of our information, and the fairness of our social systems. So when the CEO of the world’s most influential AI company says to expect some bad things, my first question is: who gets to define ‘some’?

Why does this feel so unsettling?

It’s not the concept of a tradeoff that bothers me. I get it. Progress always has a cost. Cars create pollution and cause fatal accidents. The internet gave us unprecedented access to information and also to harassment, misinformation, and scams. Every powerful technology is a package deal. We accept these tradeoffs because we’ve built systems around them over decades—seatbelts, emissions standards, insurance, digital literacy, content moderation. We have laws, regulations, and social norms that create guardrails.

What feels unsettling about Altman’s statement is the casual glossing over of what those guardrails for AI might be. Who is ‘we’ in the sentence ‘we should accept bad things’? Is it the engineers in San Francisco? Is it the regulators in Washington D.C.? Or is it me, trying to figure out if a job opportunity is real or a sophisticated phishing attempt powered by a large language model? Is it my artist friend who just saw her portfolio’s value tank because a client decided to use Midjourney instead? The ambiguity is the point. It’s a blank check.

When we introduced the automobile, we didn’t just accept a certain number of deaths as the cost of doing business and call it a day. We spent a century developing traffic laws, driver’s education, and crumple zones. With AI, the technology is moving so fast that we’re barely having the conversation about what a speed limit should even look like. Sam Altman’s comment feels like an attempt to preempt that conversation by normalizing failure before we’ve even agreed on the rules of the road. It feels like we’re being asked to sign a user agreement for society without getting to read the terms and conditions.

What could ‘some bad things’ actually look like?

Let’s get specific. Because ‘bad things’ is a category, not a reality. My fear isn’t a killer robot from a movie. My fears are smaller, more personal, and a lot more plausible.

A ‘bad thing’ is a hiring algorithm, trained on decades of biased corporate data, that consistently ranks resumes with female-sounding names lower. It’s not malicious. It’s just code reflecting the world it was shown. The harm is real. But it’s almost impossible to prove on an individual level. You just didn’t get the interview. That’s it.

A ‘bad thing’ is my local community Facebook group being flooded with AI-generated comments and fake profiles an election cycle. The goal is to sow just enough confusion and anger to depress voter turnout. The posts look real. The arguments seem plausible. The damage is done quietly, one angry comment at a time.

A ‘bad thing’ is a creative professional seeing their entire career, a craft honed over 20 years, devalued overnight. Not because an AI is more creative, but because it’s cheaper and faster for a client who only needs ‘good enough.’ This isn’t a hypothetical. I see this anxiety in my inbox and in my feeds every single week from photographers, writers, and illustrators.

A ‘bad thing’ is an AI-powered diagnostic tool that works wonderfully for 90% of the population but has a statistically higher error rate for a specific minority group due to biased training data. For the hospital system, the aggregate numbers look great. A huge win. For the person who gets the misdiagnosis, it’s a catastrophe. Who is responsible for that error? The doctor who trusted the tool? The hospital that bought it? The company that built it?

These are the kinds of ‘bad things’ I worry about. They are not cinematic explosions. They are quiet erosions of trust, opportunity, and fairness. They are statistical harms that are easy to dismiss as outliers until they happen to you.

What should I do about it?

This isn’t a call to throw our laptops in a river. That’s not happening. This technology is here, and as Altman correctly notes, its potential for good is immense. I use it every day. It helps me brainstorm, code, and summarize dense reports. The answer isn’t to reject it. The answer is to engage with it critically.

First, I have to be a skeptical user. I can’t treat ChatGPT or any other AI as an oracle. I have to question its outputs, check its facts, and be constantly aware of its potential for hallucination and bias. The ‘copy-paste’ workflow is a trap. It trades a few minutes of convenience for an abdication of critical thought. I refuse to make that trade.

Second, I can vote with my attention and my wallet. I’m actively seeking out tools and companies that are transparent about their training data, their limitations, and their safety policies. It’s getting harder to find them, but they exist. I want to support the Anthropics of the world, not because they are perfect, but because they are at least centering the conversation on safety, not just capability.

Third, I need to pay attention to the boring stuff. AI regulation sounds terribly dull. It’s full of acronyms and policy debates. But this is where the lines are being drawn. This is where we, as a society, decide what ‘some bad things’ means. Following the efforts around the EU’s AI Act or the discussions happening in Congress is no longer optional for me. It’s part of my responsibility as a citizen living in this new reality.

Ultimately, Sam Altman is not wrong that there will be tradeoffs. But framing it as something we must simply accept is a rhetorical move designed to speed past the deliberation. My job, and I think yours too, is to slow down and ask the questions. What are the specific harms? Who will they affect most? And what systems of redress and accountability will we build for them? Accepting ‘bad things’ is easy. Building a better, safer future is hard work.

FAQ

Q: Did Sam Altman give any examples of the ‘bad things’ we should accept? A: No, the comments reported were general. The lack of specifics is a major reason for the concern, as it leaves the door open to a very wide range of potential harms being considered ‘acceptable.’

Q: What is the main difference between OpenAI’s and Anthropic’s approach to AI safety? A: Anthropic has built its brand on a safety-first approach, emphasizing techniques like ‘Constitutional AI’ to align models with human values from the start. OpenAI, while having safety teams, appears to be pursuing a strategy of rapid capability scaling and deployment, with CEO Sam Altman advocating for a ‘lighter-touch’ on regulation to encourage innovation, accepting more risk as a result.

Q: Is AI going to take my job? A: AI will more likely change your job than take it entirely. Some tasks will be automated, but it will also create new needs for people who can work with AI tools, manage them, and interpret their outputs. The key is to focus on skills that complement AI, like critical thinking, creative problem-solving, and strategy.

Q: As an individual, what’s the most important thing I can do? A: Develop your own AI literacy. Don’t just be a passive user. Understand the basic concepts of how these models work, their common failure points (like bias and hallucination), and be critical of the information they provide. This empowers you to use the tools effectively without being controlled by them.