Hey y’all,
A quick thought from the weekend — that Facebook AI Policy report from Reuters (covered in the Friday issue but also covered seemingly everywhere) sure created a lot of conversation. I’m starting to have mixed feelings.
As a dad of two young guys who mean the world to me, reading those policy specifics made my stomach churn. No doubt.
But I do wonder what context (from the leaked policies) we’re missing from Reuters and how much bias against Facebook/Meta/Zuck in general is playing a role here.
I have my own sentiments on that front, sure. But there’s a necessary conversation to be had about AI’s nature (we’re anthropomorphizing a lot now) and our own acceptance of that nature.
Facebook’s policies hint at this but we need more to frame a good public discourse. As happens so often with social-media related things, we have a good headline but seem to lack the facts.
Chase the buzz,
Clay
Today’s story
“Anthropic is giving AI its own wellness protection”
My favorite weekend story came Friday from Anthropic.
In this newest research report, the lab announced that its Claude Opus 4 models can now terminate conversations if they may harm the model’s ‘moral welfare’. The firm frames this not just as user safety, but as protecting the model’s from manipulation or harm itself. Key points:
The experiment is tied to Anthropic’s questions around the consciousness and “experience” of models, themselves: Anthropic emphasizes “We remain highly uncertain about the potential moral status of Claude and other LLMs” right off the bat, but that this new feature is part of that exploration. Go deeper here.
AI morality and autonomy: Anthropic says its models will first try to redirect the user, but if extreme violations continue they will state that the interaction is over and close the chat. Claude is not allowed to do this when people express self‑harm or threats. Anthropic emphasizes their continued position of human welfare being central to their solutions.
Regulatory implications: Building the “right to exit” into models raises questions about AI rights and responsibilities. Are we moving toward a world where AI has an ethical duty to itself? How will regulators treat a system that can refuse tasks? I have a hunch that the bias topic is going to rear its head again soon.
Back to Facebook for a moment… : When an AI chooses to withdraw from abusive interactions, it mirrors boundaries that people set. The story reflects our own interpersonal norms, but so does the policy leak at Facebook which points to weird — even offensive — behavior, but behavior which is not actually illegal (yet).
Interesting human/AI news
This Past Weekend
Everybody seems willing to break company policy to use AI
Technically this report is 5 days old… News to us / probably news to you. Calypso AI, an AI security platform based in NYC & Dublin, reported in slurry of troubling figures around how employees across industries (regulated too) ignore security standards to leverage AI. More than half (52%) of us do it.AI avatars annoy some within the fashion industry
A PBS report looked at how AI‑generated models are infiltrating fashion marketing campaigns like a high profile recent campaign at Vogue. It’s an interesting debate about representation, respecting long-standing industry norms and culture.Cherokee Nation pioneers culturally‑rooted AI governance
Good piece by Forbes over the weekend. Here, the writer covers the Cherokee Nation’s CIO, Paula Starr and her position on AI policies aligned to more than efficiency gains and creativity. A compelling line: “The concept of ROI here stretches to include aspects of citizen trust, cultural preservation, and legal autonomy.” It’s going to get interesting as we ask more and more “Does using this AI align to our cultural/religious/community standards?”Regulators probe Meta’s chatbot policies
Looks like Senator Josh Hawley is launching more investigation into Meta’s leaked policies allowing generative chatbots to engage in romantic or sexual language with minors. With the EU’s AI Act and U.S. Kids Online Safety Act looming, how will we balance compliance, morality and “move fast and break things” thinking?“Ethical AI” as marketing spin
Love a provocative Medium essay. Good stuff here by Sam Liberty, arguing that “ethical AI” is akin to the new “clean coal.” Is it real, is it marketing spin, and do we know what it really means? As Sam puts it, the real test is how company ethics and principles persist when business pressure mounts up against them.More languages get speech AI access, thanks to Nvidia
Nvidia’s release of its Granary dataset (around one million hours of multilingual speech data) as well as open models (named Canary and Parakeet) unlock accurate transcription and translation solutions for 25 under-served European languages. (I had no idea!) It’s good to see progress here, protecting language and cultural diversity in the age of AI.
I'd love to hear from you!
Thanks as always, human.
