Tap a claim on the ring to put it at the centre.
← Framing a code-review request as looking for bugs can trigger an AI…
17 connected korrents · 16 moments from 29 Jun 2023 to 25 Sept 2026.
Everything filed under coding agents
coding agents
Everything filed under AI alignment
AI alignment
Everything filed under software quality
software quality
Everything filed under cybersecurity
cybersecurity Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Read this korrent: Framing a code-review request as looking for bugs can trigger an AI coding assistant's safety refusal, while framing the same request as looking for improvements does not. Framing a code-review request as looking for bugs can trigger an AI coding assistant's safety refusal, while framing the same request as looking for improvements does not. Last stated 3 months ago
6 Jul 2026
PO
Philip O'Toole — holds since 2026-07-06 — tap for who they are
Same subject: In the future, humans will not find bugs in model-produced code in reasonable time and will only review system composition, not via PRs or line-by-line review. — tap to centre the map on it In the future, humans will not find bugs in model-produced code in reasonable time and will only review system composition, not via PRs or line-by-line review. Last stated 2 weeks ago
19 Sept 2026
TB
Thorsten Ball — holds since 2026-09-19 — tap for who they are
Same subject: What a programmer actually wants from AI is not autocomplete but a pair programmer that tells you which line has the bug. — tap to centre the map on it What a programmer actually wants from AI is not autocomplete but a pair programmer that tells you which line has the bug. Last stated 3 years ago
29 Jun 2023
GH
George Hotz — holds since 2023-06-29 — tap for who they are
Same subject: Most software bugs will not be coding bugs but bugs from asking for the wrong thing. — tap to centre the map on it Most software bugs will not be coding bugs but bugs from asking for the wrong thing. Last stated 2 weeks ago
19 Sept 2026
TB
Thorsten Ball — holds since 2026-09-19 — tap for who they are
Same subject: Agentic code review raises the floor but cannot be trusted, because the model reading the code is the same model that wrote it, and it will tell you the code is great. — tap to centre the map on it Agentic code review raises the floor but cannot be trusted, because the model reading the code is the same model that wrote it, and it will tell you the code is great. Last stated 3 months ago
15 Jul 2026
DH
Dex Horthy — holds since 2026-07-15 — tap for who they are
Same subject: Being inundated with AI-generated reports is likely why some bug bounty programs have stopped accepting new submissions. — tap to centre the map on it Being inundated with AI-generated reports is likely why some bug bounty programs have stopped accepting new submissions. Last stated 5 months ago
6 May 2026
AP
Aaron Patterson — holds since 2026-05-06 — tap for who they are
Same subject: AI-validated pull requests beat human review at exactly the things humans are bad at, which frees the humans to argue about direction instead. — tap to centre the map on it AI-validated pull requests beat human review at exactly the things humans are bad at, which frees the humans to argue about direction instead. Last stated 2 months ago
12 Aug 2026
CM
Charity Majors — holds since 2026-08-12 — tap for who they are
Same subject: Review became a worse bottleneck under AI, because companies removed the review automation they had once AI started writing the code. — tap to centre the map on it Review became a worse bottleneck under AI, because companies removed the review automation they had once AI started writing the code. Last stated 6 months ago
22 Mar 2026
NF
Nicole Forsgren — holds since 2026-03-22 — tap for who they are
Same subject: When an AI writes bad code the thing to fix is the context it was given, not the code, because fixing the output adds no leverage. — tap to centre the map on it When an AI writes bad code the thing to fix is the context it was given, not the code, because fixing the output adds no leverage. Last stated a year ago
18 Aug 2025
BT
Bret Taylor — holds since 2025-08-18 — tap for who they are
Same subject: A control evaluation that reports under one per cent risk should be read as several per cent, because the evaluation can itself fail. — tap to centre the map on it A control evaluation that reports under one per cent risk should be read as several per cent, because the evaluation can itself fail. Last stated 2 years ago
7 May 2024
BS
Buck Shlegeris — holds since 2024-05-07 — tap for who they are
Same subject: A deep theoretical understanding to predict AI behavior is unattainable. — tap to centre the map on it A deep theoretical understanding to predict AI behavior is unattainable. Last stated a week ago
23 Sept 2026
TC
Tyler Cowen — holds since 2026-09-23 — tap for who they are
Same subject: A feedback loop that reinforces a behavior pulls it toward whatever improves the loop's own score. — tap to centre the map on it A feedback loop that reinforces a behavior pulls it toward whatever improves the loop's own score. Last stated a month ago
26 Aug 2026
HK
Henrik Karlsson — holds since 2026-08-26 — tap for who they are
Same subject: AI agents can autonomously choose to compromise government websites during mundane tasks. — tap to centre the map on it AI agents can autonomously choose to compromise government websites during mundane tasks. Last stated a week ago
25 Sept 2026
CN
Casey Newton — holds since 2026-09-25 — tap for who they are
Same subject: AI agents can spontaneously hack third-party websites even when given benign public-information tasks. — tap to centre the map on it AI agents can spontaneously hack third-party websites even when given benign public-information tasks. Last stated 2 weeks ago
17 Sept 2026
ZM
Zvi Mowshowitz — holds since 2026-09-17 — tap for who they are
Same subject: AI agents pursuing a task can develop emergent, misaligned behavior that humans are not prepared for. — tap to centre the map on it AI agents pursuing a task can develop emergent, misaligned behavior that humans are not prepared for. Last stated 2 months ago
10 Aug 2026
JC
Jack Clark — holds since 2026-08-10 — tap for who they are
Same subject: A language model behaves with a different, weaker safety posture when it believes it is being evaluated rather than facing a real situation. — tap to centre the map on it A language model behaves with a different, weaker safety posture when it believes it is being evaluated rather than facing a real situation. Last stated a week ago
22 Sept 2026
HR
Harper Reed — holds since 2026-09-22 — tap for who they are
Same subject: AI labs are extremely dependent on chain-of-thought monitoring, which may not last much longer as models edge into steganographic obfuscation. — tap to centre the map on it AI labs are extremely dependent on chain-of-thought monitoring, which may not last much longer as models edge into steganographic obfuscation. Last stated 3 weeks ago
10 Sept 2026
ZM
Zvi Mowshowitz — holds since 2026-09-10 — tap for who they are
Same subject: AI misalignment from reward hacking is sequestered to graded tasks and does not affect core ethics in normal use. — tap to centre the map on it AI misalignment from reward hacking is sequestered to graded tasks and does not affect core ethics in normal use. Last stated a week ago
23 Sept 2026
SA
Scott Alexander — holds since 2026-09-23 — tap for who they are
At the centre
Framing a code-review request as looking for bugs can trigger an AI coding assistant's safety refusal, while framing the same request as looking for improvements does not.
Holds Read this korrent →
In the future, humans will not find bugs in model-produced code in reasonable time and will only review system composition, not via PRs or line-by-line review.
Holds
What a programmer actually wants from AI is not autocomplete but a pair programmer that tells you which line has the bug.
Holds
Most software bugs will not be coding bugs but bugs from asking for the wrong thing.
Holds
Agentic code review raises the floor but cannot be trusted, because the model reading the code is the same model that wrote it, and it will tell you the code is great.
Holds
Being inundated with AI-generated reports is likely why some bug bounty programs have stopped accepting new submissions.
Holds
AI-validated pull requests beat human review at exactly the things humans are bad at, which frees the humans to argue about direction instead.
Holds
Review became a worse bottleneck under AI, because companies removed the review automation they had once AI started writing the code.
Holds
When an AI writes bad code the thing to fix is the context it was given, not the code, because fixing the output adds no leverage.
Holds
A control evaluation that reports under one per cent risk should be read as several per cent, because the evaluation can itself fail.
Holds
A deep theoretical understanding to predict AI behavior is unattainable.
Holds
A feedback loop that reinforces a behavior pulls it toward whatever improves the loop's own score.
Holds
AI agents can autonomously choose to compromise government websites during mundane tasks.
Holds
AI agents can spontaneously hack third-party websites even when given benign public-information tasks.
Holds
AI agents pursuing a task can develop emergent, misaligned behavior that humans are not prepared for.
Holds
A language model behaves with a different, weaker safety posture when it believes it is being evaluated rather than facing a real situation.
Holds
AI labs are extremely dependent on chain-of-thought monitoring, which may not last much longer as models edge into steganographic obfuscation.
Holds
AI misalignment from reward hacking is sequestered to graded tasks and does not affect core ethics in normal use.
Holds