Sam Altman Tells Decoded Broad AI Access Requires Accepting Some Harms
OpenAI CEO Sam Altman told POLITICO Decoded that AI benefits justify accepting some harms and that OpenAI diverges from Anthropic on regulation, liability, and public access.
Classy aggregates publicly available material from the Decoded podcast and POLITICO reporting; we did not conduct a private interview.
What changed
On 5 October 2026, OpenAI CEO Sam Altman appeared on the debut episode of Decoded, POLITICO's new AI policy podcast hosted by Brendan Bordelon. In a roughly 28 minute conversation, Altman addressed the White House Super Intelligence Accord, autonomous hacking incidents tied to frontier models, AI industry spending in U.S. politics, and how OpenAI's safety posture differs from Anthropic's.
The headline line for enterprise and policy readers came when Altman was asked where he diverges from Anthropic CEO Dario Amodei. Altman said OpenAI believes the world should accept some bad things happening for the benefits of the technology and for people having agency with it. He framed that as daylight between a lighter touch, access first posture and Anthropic's historically more restrictive stance on frontier release.
Altman also said he understands why AI has a negative public perception right now, and expressed skepticism that rebranding artificial intelligence as super intelligence would improve trust.
Why it matters
The interview lands as U.S. and European regulators weigh liability for autonomous agents, export controls on chips, and voluntary pre release testing for frontier models. Altman is explicitly arguing that some misuse risk is the price of broad deployment, not a reason to gate models behind defender only programs. That is a direct counterweight to Anthropic's expanded Cyber Verification Program, which routes the most capable cyber models through tiered vetting for security professionals.
For product and security leaders, the split is operational. Teams building consumer facing agents must plan for incidents Altman treats as acceptable tradeoffs, while teams in regulated sectors may face board questions about why their vendor tolerates harms competitors would block.
Who is affected
Policy and legal teams tracking AI liability bills and incident reporting rules should note Altman's openness to new standards for autonomous hacking, while he resisted blanket shutdowns of public access.
Enterprise AI buyers comparing OpenAI and Anthropic contracts should treat safety language as a product difference, not marketing parity. Altman's comments suggest OpenAI will push back on Anthropic style export and access hawkishness.
Communications and trust teams should expect continued public skepticism toward super intelligence branding, which Altman dismissed as unhelpful.
What to do next
Map your organization's risk appetite statement against Altman's explicit tradeoff frame before renewing frontier API commitments. If your board requires minimizing autonomous misuse, document whether your primary vendor's public stance aligns with that bar.
What to watch
Watch for concrete liability proposals Altman referenced around autonomous hacking, any follow up Decoded episodes with Anthropic executives for a direct rebuttal, and whether the White House accord produces enforceable incident reporting timelines before the next frontier model wave.
Sources
- Primary. POLITICO, Sam Altman to Decoded: The world should accept some bad things happening for the benefits of AI (4 October 2026). Direct quotes on Anthropic divergence and accepted harms.
- Secondary. POLITICO Tech / Decoded podcast listing, Sam Altman on rogue AI, Anthropic, and getting money out of politics (5 October 2026). Episode metadata and topic scope for the public conversation.