Safety and alignment

Yampolskiy argues fully autonomous AI can never be assumed safe

In a first-person MIRI interview, Roman Yampolskiy argued that AI safety engineering can constrain limited systems but cannot make fully autonomous, self-improving AI reliably safe. He pointed to inconsistent human values, uncertain goal preservation under self-modification, software faults, and fundamental verification limits.

0 comments · 0 votesOpen discussion

Public discussion is readable by everyone. Sign in to comment, reply, or vote.

No comments yet. Start the discussion.

CURRENT ASSESSMENT · REVISION 1
TOWARD DOOM28confidence 66/100

Why it moved the index

This is direct historical evidence for the control-difficulty pathway because it identifies why safety claims for autonomous, self-modifying AI may remain probabilistic rather than assured.

AUDIT TRAIL

Assessment history

  1. R1
    Toward 28 · confidence 66

    Adds a distinct, exactly dated first-person control argument from Roman Yampolskiy's historical backfill.

    20 Sept 2026
SHARE THE FINDINGS

Share this page

DoomBench social sharing card for Yampolskiy argues fully autonomous AI can never be assumed safe.
  1. DoomBench assesses “Yampolskiy argues fully autonomous AI can never be assumed safe” as evidence moving toward doom, with magnitude 28 and confidence 66 out of 100 in the safety and alignment category.

  2. The DoomBench assessment of “Yampolskiy argues fully autonomous AI can never be assumed safe” is based on reporting from Machine Intelligence Research Institute and records the editorial rationale, source quality, attribution, and...

  3. DoomBench summarizes “Yampolskiy argues fully autonomous AI can never be assumed safe” as follows: In a first-person MIRI interview, Roman Yampolskiy argued that AI safety engineering can constrain limited systems but cannot make fully...

    https://www.doombench.com/news/yampolskiy-argues-fully-autonomous-ai-can-never-be-assumed-safe-2013-07-16