Ted Chiang: Stop pretending the chatbot has feelings

The sci-fi writer delivers the clearest takedown yet of AI anthropomorphism, calling it a corporate con that shifts blame from builders to bots.

Why it matters: As Anthropic publishes 84-page "constitutions" for Claude and Dario Amodei muses about model consciousness, Chiang argues the whole frame is wrong. Treating LLMs as moral agents lets companies off the hook when the tech causes harm.

• The core argument: a chatbot is sentence completion dressed up as dialogue. The "Julius Caesar character" in an LLM prompt isn't conscious. Neither is the "helpful AI chatbot" character. Same mechanism, same non-experience.
• The danger: If we accept that AI has feelings or moral agency, we stop holding the companies that built and deployed it responsible for what it does.

Go deeper: https://www.theatlantic.com/philosophy/2026/06/no-artificial-intelligence-is-not-conscious/687378/

AI attackers are harder to spot after a year of data

Anthropic studied 832 banned accounts over 12 months and found AI is making lower-skilled threat actors significantly more dangerous, fast.

Why it matters: The share of medium-or-higher-risk actors jumped from 33% to 56% in just six months. The old signals security teams use to triage threats no longer work when AI can perform advanced techniques on behalf of anyone.

• AI use is shifting deeper into attacks: phishing (initial access) fell 8.6% while account discovery inside compromised networks rose 8.9%. Attackers are going post-compromise.
• MITRE ATT&CK has no category for agentic orchestration, the thing that made a state-sponsored espionage op earn a risk score of 100 in Anthropic's system despite looking medium-risk by technique count alone.

Go deeper: https://www.anthropic.com/news/AI-enabled-cyber-threats-mitre-attack

Anthropic formalizes its enterprise partner tiers

Anthropic launched a tiered Services Track and public Partner Hub for the Claude Partner Network, giving enterprises a structured way to find qualified implementation firms.

Why it matters: More than 40,000 firms applied to join the partner network since its March launch. The new tiers (Select, Preferred, Global Premier) put real requirements on firm size and production deployments, not just revenue sent to Anthropic.

• The biggest consultancies are all in: Accenture (30,000 trained), Deloitte (470,000 people), Cognizant (350,000 associates), PwC rolling out Claude Code globally.
• Tier standing is public and refreshed daily in the Partner Hub, meaning customers can verify a firm's certified headcount and production deployments before signing.

Go deeper: https://www.anthropic.com/news/services-track-partner-hub