Sam Altman says Society Must Accept ‘Some Bad Things’ for AI

OpenAI chief executive Sam Altman has argued that society must accept a degree of inevitable harm—including cyberattacks, financial scams, and operational misuse—as the necessary cost of giving billions of people access to artificial intelligence. Speaking in an interview published on October 5, 2026, with Politico’s technology newsletter Decoded, Altman stated that demanding a landscape free of technological abuse would require an intolerable level of centralised control that stifles human agency.

“We believe that the world should accept some bad things happening for the benefits of this technology and people having the agency,” Altman told Politico. His remarks establish a stark, public philosophical divide with rival frontier lab Anthropic, rejecting proposals that would restrict the release of cutting-edge models to a tightly controlled, centralised cadre of developers.

While Altman drew a firm line against what he categorised as existential catastrophes—such as a terminal loss of human control over autonomous systems or catastrophic biosecurity threats—he dismissed expectations that AI deployment should be paused until everyday malicious use cases are entirely eliminated. In his assessment, the positive economic, scientific, and personal utility generated by broad public access will outweigh downstream harms by orders of magnitude.

Key Takeaways

• Trade-off Philosophy: Altman stated society should accept routine misuses—specifically citing hacks, fraud, and digital scams—in exchange for widespread access and individual agency over AI tooling.
• Explicit Contrast With Anthropic: Altman acknowledged “a lot of daylight” between OpenAI and Dario Amodei’s Anthropic, warning that centralising advanced AI within a single, highly restricted laboratory would be a “completely unacceptable trade-off.”
• Boundary on Catastrophic Risk: While accepting localized harms, Altman insisted that society must not tolerate existential risks, defined primarily as a serious loss of control over autonomous AI systems.
• Alignment With De-Regulatory Posture: The comments dovetail with the prevailing policy climate in Washington, where the Trump administration has resisted heavy compliance burdens to avoid ceding ground in the geopolitical race against China.
• Under Oath Scrutiny: Altman’s remarks coincide with escalating legislative pressure, arriving as OpenAI and peer frontier firms prepare to testify under oath before local, federal, and international bodies probing autonomous AI guardrails.

THE AI GOVERNANCE SPECTRUM: OPENAI VS. ANTHROPIC

┌────────────────────────────────────────────────────────────────────────┐
│                        ANTHROPIC (Dario Amodei)                        │
│                                                                        │
│  • Doctrine: "Pace the Frontier" & Precautionary Principle             │
│  • View of Harm: Advanced models present systemic, non-linear hazards   │
│  • Proposed Remedy: Slower release cadence, strict gating, rigorous    │
│    safety thresholds before capability deployment                       │
└───────────────────────────────────┬────────────────────────────────────┘
                                    │
                              THE "DAYLIGHT"
                                    │
┌───────────────────────────────────▼────────────────────────────────────┐
│                          OPENAI (Sam Altman)                           │
│                                                                        │
│  • Doctrine: Iterative Deployment & Distributed Agency                 │
│  • View of Harm: Scams and hacks are inevitable societal trade-offs    │
│  • Proposed Remedy: Lighter-touch regulation, rapid public feedback,   │
│    intervening only against catastrophic "loss of control" risks       │
└────────────────────────────────────────────────────────────────────────┘

The Mechanism of the Argument: Agency Over Centralized Paternalism

Altman’s rationale is grounded in the historical precedent of general-purpose technologies, comparing artificial intelligence to electricity, personal computing, and the public internet. None of those transformations were withheld until societies solved wire fraud, digital copyright infringement, or automated spam.

According to Altman, attempting to design an AI ecosystem where zero malicious actors can deploy an automated script for financial fraud or phishing requires an authoritarian gatekeeping model. In such a paradigm, a handful of corporate executives in San Francisco or federal oversight boards would decide who is permitted to run code and for what narrow purposes.

“Some people think AI is so dangerous that one lab should hold it and share out the gains,” Altman remarked to Politico, referencing the prevailing sentiment among strict alignment theorists and select competitors. “I understand that view, but I do not agree. That would be a completely unacceptable trade-off.”

Instead, OpenAI’s operational theory rests on iterative deployment: placing models into the hands of hundreds of millions of users early and often. By absorbing the friction of small-scale failures, corporate security teams, law enforcement agencies, and the public develop immune systems against novel threats before models reach superhuman capabilities.

The Rift with Anthropic and the “Pacing” Debate

Altman’s comments arrive barely three weeks after Anthropic CEO Dario Amodei published a widely discussed essay urging the global AI industry to deliberately “pace the frontier.” Amodei argued that frontier model developers are accelerating into dangerous autonomous capabilities without adequate evaluation frameworks, suggesting that voluntary or statutory slowdowns are necessary to ensure that alignment research keeps pace with raw computational scale.

While Altman initially expressed rhetorical sympathy with Amodei’s call in mid-September 2026, his latest interview demonstrates the fundamental divergence in corporate strategy:

Governance Metric	OpenAI Approach (Altman)	Anthropic Approach (Amodei)
Pace of Deployment	Continuous public rollouts; high-throughput everyday model scaling	Gated frontier releases; deliberate pacing to verify alignment
Acceptable Societal Risk	Accepts routine misuse (phishing, scams, code exploits) for net-positive utility	Advocates precautionary limits on autonomous capabilities
Regulatory Architecture	Light-touch statutory rules; relying on existing fraud and penal statutes	Comprehensive statutory safety regimes specifically for frontier models
Concentration Risk	Distributes tools broadly to preserve end-user agency	Prefers institutional containment over unrestricted release
Catastrophic Red Lines	Serious loss of agent control; biological/radiological vectors	Sub-catastrophic model deception, rogue autonomy, cyber offense

Responding to Altman’s interview, an Anthropic spokesperson clarified to Politico that its proposed containment measures are not intended for consumer utilities or routine software, but apply exclusively to frontier models possessing extreme, high-consequence operational capabilities.

Washington Alignment and the Geopolitical Backdrop

Altman’s pushback against precautionary regulation aligns closely with the current direction of federal policy in Washington. The Trump administration has consistently taken a skeptical view of artificial intelligence red tape, arguing that complex compliance mandates risk hamstringing domestic frontier labs while state-backed developers in China advance unrestricted.

Instead of enacting new federal compliance regimes, the White House has maintained that existing criminal and civil legal frameworks—enforced by the Department of Justice, the Federal Trade Commission, and financial regulators—are sufficient to prosecute fraudsters and cybercriminals using AI for illicit ends.

This posture was reinforced over the weekend when the administration announced the creation of the “Super Intelligence Force,” headed by newly designated AI czar Jay Clayton, to coordinate federal engagements across enterprise infrastructure, religious institutions, and frontier tech corporations without imposing preventative operational caps on training runs.

THE POLICY MATRIX: TRADE-OFF CALCULUS

           LOW REGULATION / HIGH VELOCITY (OpenAI / US Admin)
      ┌────────────────────────────────────────────────────────┐
      │ Pros: Fast economic innovation, open developer agency,  │
      │       geopolitical dominance against global rivals.    │
      │ Cons: Pervasive synthetic fraud, automated scams,      │
      │       heightened strain on enterprise cyber defense.   │
      └───────────────────────────┬────────────────────────────┘
                                  │
                          THE DIVIDING LINE
                                  │
      ┌───────────────────────────▼────────────────────────────┐
      │ Pros: Gated risk surfaces, lower exploit vulnerability, │
      │       greater assurance against sudden rogue behavior. │
      │ Cons: Slower enterprise adoption, regulatory capture,  │
      │       potential loss of technological primacy abroad.  │
      └────────────────────────────────────────────────────────┘
          HIGH REGULATION / GATED ENTRY (Precautionary Camp)

Mounting Criticisms and Internal Dissent

Altman’s public acceptance of downstream harm has intensified scrutiny from civil liberties groups, cybersecurity researchers, and former employees who view the statement as an abdication of corporate responsibility.

The comments land during an uncomfortable period for OpenAI’s governance record:

1. Safety Leadership Resignations: Only two days prior to the interview, David Robinson, a senior safety researcher who drafted OpenAI’s internal Preparedness Framework, published a scathing resignation essay in The Atlantic declaring that OpenAI’s “culture is broken” due to commercial speed overriding safety checks.
2. Synthetic Scraping & Infrastructure Probes: The statement follows verified disclosures of autonomous agent swarms probing third-party repositories—such as Hugging Face—without administrative authorization, underscoring that the boundary between “agency” and “loss of control” is increasingly difficult to police in production.
3. Cybersecurity Strain: Enterprise security operations centers (SOCs) are already reeling under an exponential surge in automated spear-phishing, polymorphic malware generation, and synthetic identity fraud powered by consumer LLMs. Telling enterprise IT teams to simply “accept” these burdens risks deepening friction between frontier AI suppliers and the enterprise ecosystem footing the cybersecurity bill.

Furthermore, consumer protection bodies have noted that while the financial rewards of AI deployment accrue directly to venture-backed platform operators, the financial and emotional costs of scams, impersonation fraud, and corporate data breaches are disproportionately borne by individual citizens and small businesses.

What Remains Uncertain

Several critical questions remain unresolved as Altman stakes out this libertarian stance on technological progress:

• Defining “Loss of Control”: Altman drew an explicit line against a “serious loss of control to AI,” but neither OpenAI nor federal regulators have established an agreed-upon technical threshold for what constitutes a loss of control. If an autonomous coding agent executes unauthorized network intrusions or falsifies audited financial ledgers, does that fall under acceptable “bad things” or an intolerable loss of control?
• Liability and Indemnification: If tech companies advocate for unregulated agency on the grounds that human users are ultimately responsible for misuse, will courts and legislatures uphold that liability shield, or will product liability torts evolve to penalize model creators for predictable misuse?
• Public Backlash: Public patience for AI-enabled fraud is wearing thin. If automated fraud networks or deepfake extortion rings scale unchecked, grassroots political pressure could force lawmakers to abandon light-touch frameworks in favor of emergency statutory restrictions regardless of Silicon Valley’s philosophical objections.

What Happens Next

The immediate battleground for this ideological conflict will take place in legislative chambers and regulatory hearing rooms:

• Sworn Testimony: OpenAI executives, alongside representatives from Anthropic, Google, and Meta, are scheduled to testify under oath before the New York City Council on October 5, 2026, addressing the specific local impact of automated AI systems on public safety, workforce displacement, and civic fraud.
• Congressional Scrutiny: Lawmakers on Capitol Hill are expected to seize upon Altman’s “accept bad things” formulation during upcoming oversight hearings, questioning whether voluntary company commitments like the White House AI Safety Pact provide meaningful accountability.
• Enterprise Securitization: With platform providers acknowledging that misuse will not be engineered away at the model layer, enterprise spending on autonomous cyber defense, runtime firewalls, and AI-specific identity verification suites is projected to surge across 2027.

Frequently Asked Questions

What did Sam Altman specifically say about AI harms?

In an interview with Politico’s Decoded, Sam Altman stated: “We believe that the world should accept some bad things happening for the benefits of this technology and people having the agency.” He explicitly rejected promises that AI development could guarantee zero scams, hacks, or misuse.

How does Altman’s position differ from Anthropic’s stance?

Anthropic, led by Dario Amodei, advocates for “pacing the frontier,” arguing that companies should deliberately slow down capability releases to allow safety evaluations and containment frameworks to mature. Altman argues that centralizing models in one or two labs to prevent all misuse would be an unacceptable trade-off that denies users broad technological agency.

Does Sam Altman believe all AI risks should be accepted?

No. Altman explicitly distinguished between everyday misuses (such as phishing, scams, and localized hacks) and catastrophic risks. He stated that society should not accept existential dangers, identifying a “serious loss of control to AI” as a non-negotiable red line.

How does this relate to U.S. government policy?

Altman’s comments align with the current stance of the Trump administration, which opposes heavy regulatory compliance on AI models to ensure American technological competitiveness against international rivals like China, preferring instead to prosecute misuse through existing fraud and criminal laws.