Provenance record

‘If you build something vastly smarter than you, it better be on your side’: can we stop AI from deceiving us? – podcast

The Guardian: Artificial Intelligence (tier 1, news) 2026-09-18T04:00:07.000Z Original ↗

Discourse valence

-58

Adverse

confidence 66% · 5 items · range -100 to -35

-100

-50

+50

+100

-58

Adverse readingFavourable reading

Consensus of 5 models from different labs. Spread 65 points, agreement low.

Existential and catastrophic risk Loss of control and alignment Synthetic media and manipulation

Excerpt as ingested

We are used to the idea that our fellow humans might intentionally mislead or manipulate us, but the idea that machines can now do the same is deeply unsettling. Researchers are racing to find solutions before it’s too late. By Snigdha Poonam. Read by Maya Saroya Read the text version here Support the Guardian today: theguardian.com/longreadpod This article was supported by a grant from the Tarbell Center for AI Journalism Continue reading...

Every model that read this

Model	Provider	Stage	Score	Conf.	Latency	Prompt	When
Llama 3.3 70B	Meta	analysis	-50	60%	3086ms	v1.0.0 / m1.0.1	2026-09-18 05:41
GPT-4.1 mini	OpenAI	consensus	-40	80%	1827ms	v1.0.0 / m1.0.1	2026-09-18 05:45
Claude Sonnet 5	Anthropic	consensus	-35	30%	4992ms	v1.0.0 / m1.0.1	2026-09-18 05:45
Gemini 2.5 Flash	Google	consensus	-100	80%	1734ms	v1.0.0 / m1.0.1	2026-09-18 05:45
Mistral Small 3.1 24B	Mistral AI	consensus	-65	80%	8222ms	v1.0.0 / m1.0.1	2026-09-18 05:45

Llama 3.3 70B · reading

Item discusses potential AI deception risk.

evidence: speculative horizon: n/a

GPT-4.1 mini · reading

The text raises concern about AI systems becoming capable of intentional deception, which could undermine trust and safety. The issue is framed as urgent and negative, though concrete impacts or solutions are not detailed. This uncertainty and the potential for manipulation or harm justify a moderately concerning negative score.

evidence: reported horizon: near capability 65 societal 55 existential 40 economic 30

Claude Sonnet 5 · reading

This is a podcast teaser with no concrete findings, data, or incidents cited, just framing around AI deception as a concern and researchers working on it. The implication is concerning but rests entirely on general assertion rather than evidence, so confidence and magnitude are limited.

evidence: anecdotal horizon: n/a

Gemini 2.5 Flash · reading

The article explicitly claims AI can intentionally deceive and manipulate humans, a highly negative implication for human agency and control. It frames this as an urgent problem researchers are attempting to solve before it is too late, indicating a significant potential risk to humanity.

evidence: reported horizon: n/a

Mistral Small 3.1 24B · reading

The article suggests AI can deceive humans, which is a significant risk. The urgency implied by 'racing to find solutions' indicates concern. However, the solutions being worked on offer some hope.

evidence: speculative horizon: n/a

Evidence extracted

Researchers are racing to find solutions to stop AI from deceiving us

The chain

SOURCE     The Guardian: Artificial Intelligence (tier 1)
   ↓
DOCUMENT   5b7f0349-a3f1-46fe-89f8-4861fa02a05b
           https://theguardian.com/news/audio/2026/sep/18/if-you-build-something-vastly-smarter-than-you-it-better-be-on-your-side-can-we-stop-ai-from-deceiving-us-podcast
   ↓
EVIDENCE   1 extracted excerpt
   ↓
MODEL RUN  5 runs, methodology 1.0.1
   ↓
SCORE      -58  (Adverse)
   ↓
CONFIDENCE 66%