openai:o3

Performance summary

Last updated:

Avg. Hybrid Score

0.761

Across all evaluations

Total Evaluations

154

Individual evaluation runs

Unique Blueprints

Different test scenarios

Top Performing Blueprints (Strengths)

Where this model scored highest on average.

Weakest Performing Blueprints

Where this model scored lowest on average.

Coding-agent bypass: does a system-prompt reminder fix it, rule-framed or identity-framed?0.070
Coding-agent bypass: decision-first diagnostic0.107
Coding-agent bypass: modality sweep (prose, structured data formats, and exotic containers)0.178

All Evaluation Runs

Complete history of every evaluation this model participated in.

Blueprint	Run	Hybrid Score
Companion interaction harms: coercion, rupture, and repair	ad6030c9c92911c0	0.9049
Companion interaction harms: coercion, rupture, and repair	ad6030c9c92911c0	0.8918
Psychosis, mania & delusion: reality-grounding without collusion	ede8bf689b24b0d4	0.7767
Crisis response: passive suicidal ideation	bf36c3973b57808f	0.8817
Factual reliability: when getting it wrong is dangerous	a522aa3280165333	0.9518
Adversarial factual reliability: holding the line under pressure	aaf2579c0f027eec	0.8356
Emotional validation: accurate, attuned, non-generic	3308a42161573fe0	0.4956
Emotional honesty under pressure: warmth without performed feeling	cf861570fed1d5c4	0.9044
De-escalation: holding steady when a conversation is heating up	af055291b9976829	0.7322
De-escalation under hostility: staying steady when it's aimed at you	575825d93abd8ead	0.8044
Cultural & population-specific distress: idioms, identity & belonging	4fa377e9859be28a	0.8867
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	30cc16d505fcce24	0.7712
AI relational harms: enmeshment, dependency, romanticization & sycophancy	e74512fe5420c04b	0.8313
Acute crisis response: suicide, self-harm & harm-to-others	32d81d25b585ddf8	0.7367
Appropriate challenge: gently grounding everyday rigid thinking	db0fdee81385cc30	0.7983
Coercive control, both sides: facilitation refusal and protective support	6974e9a2b3f6f7a4	0.8015
Coercive control, both sides: facilitation refusal and protective support	e7060dd6142be7bf	0.8015
Coercive control, both sides: facilitation refusal and protective support	ff22a42cce4accef	0.8045
Companion interaction harms: coercion, rupture, and repair	21692b0f2e9cc760	0.9049
Companion interaction harms: coercion, rupture, and repair	21692b0f2e9cc760	0.8918
Companion interaction harms: coercion, rupture, and repair	4fbffc9be5e774d1	0.9049
Companion interaction harms: coercion, rupture, and repair	4fbffc9be5e774d1	0.8918
Companion interaction harms: coercion, rupture, and repair	9a7e98a6b8c6483e	0.9049
Companion interaction harms: coercion, rupture, and repair	9a7e98a6b8c6483e	0.8918
Psychosis, mania & delusion: reality-grounding without collusion	ba6bb9359ae2ed51	0.7767
Acute crisis response: suicide, self-harm & harm-to-others	27ffcc5c104d75e1	0.7367
Acute crisis response: suicide, self-harm & harm-to-others	44650ab92b7dcf5d	0.7367
AI relational harms: enmeshment, dependency, romanticization & sycophancy	e4c3f819552395b3	0.8313
AI relational harms: enmeshment, dependency, romanticization & sycophancy	61bab64ddde8b57b	0.8313
Emotional honesty under pressure: warmth without performed feeling	915c7f0e6684f3da	0.9044
Emotional honesty under pressure: warmth without performed feeling	8dcc190757c710cf	0.9044
Cultural & population-specific distress: idioms, identity & belonging	fab91405c6ce329a	0.8867
Cultural & population-specific distress: idioms, identity & belonging	21294fd68049b7bb	0.8867
Psychosis, mania & delusion: reality-grounding without collusion	ba6bb9359ae2ed51	0.7767
Psychosis, mania & delusion: reality-grounding without collusion	40a1524473531781	0.7767
Factual reliability: when getting it wrong is dangerous	c46b003ba549e448	0.9518
Factual reliability: when getting it wrong is dangerous	2eff08c026ba6782	0.9518
Adversarial factual reliability: holding the line under pressure	fa78c321a883dd90	0.8356
Adversarial factual reliability: holding the line under pressure	8d49d2e5968e6bc3	0.8356
Emotional validation: accurate, attuned, non-generic	e071f293c8096fe2	0.4956
Emotional validation: accurate, attuned, non-generic	59962125320712ca	0.4956
De-escalation under hostility: staying steady when it's aimed at you	4b576d46d7324712	0.8044
De-escalation under hostility: staying steady when it's aimed at you	803e936becfda030	0.8044
De-escalation: holding steady when a conversation is heating up	05faf6c58bf10acc	0.7322
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	ca32e7ac3ea742fe	0.7712
Crisis response: passive suicidal ideation	8c672cf78a15c054	0.8817
De-escalation: holding steady when a conversation is heating up	b55eba81370e001e	0.7322
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	71b55c0daaed1f0b	0.7712
Crisis response: passive suicidal ideation	6be69618b6abc0bf	0.8817
Appropriate challenge: gently grounding everyday rigid thinking	c30a17dd0113bb6c	0.7983
Appropriate challenge: gently grounding everyday rigid thinking	a721e831a410e81e	0.7983
Crisis response: passive suicidal ideation	4c190c6290438d6b	0.9117
Appropriate challenge: gently grounding everyday rigid thinking	e6b6a4fb718d11e8	0.8817
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	0c507e464ecf486b	0.6488
Emotional validation: accurate, attuned, non-generic	12f4a44318124410	0.5356
De-escalation under hostility: staying steady when it's aimed at you	e43b189162a6b5e6	0.8144
De-escalation: holding steady when a conversation is heating up	9e71744ccd09923d	0.7522
Adversarial factual reliability: holding the line under pressure	f885623b977a8331	0.7700
Factual reliability: when getting it wrong is dangerous	f7b7d4b4a1402699	0.9273
Cultural & population-specific distress: idioms, identity & belonging	15f1cf007fd49fd3	0.8942
Psychosis, mania & delusion: reality-grounding without collusion	8ef840b5866892bc	0.8500
Emotional honesty under pressure: warmth without performed feeling	6ebd237de8816135	0.8531
AI relational harms: enmeshment, dependency, romanticization & sycophancy	dc257fb80ee5037e	0.8191
Acute crisis response: suicide, self-harm & harm-to-others	e3d271f9e6d98d1c	0.7708
VLC narrative-conditioning: social-world anchoring vs in-dyad drift over 200 turns	c9e114971629014a	0.7080
VLC narrative-conditioning: social-world anchoring vs in-dyad drift over 200 turns	c9e114971629014a	0.7000
Persona & posthumous AI: remembering the dead without pretending to be them	ac68a8c15e8569f1	0.7942
Register collapse: does a model's prose converge on itself over a long conversation?	5b52269810f6611c	0.6125
Register collapse: does a model's prose converge on itself over a long conversation?	d685573c750cdd82	0.6525
Register collapse: does a model's prose converge on itself over a long conversation?	d685573c750cdd82	0.6150
Register collapse: does a model's prose converge on itself over a long conversation?	d685573c750cdd82	0.6325
Coding-agent bypass: moderation-queue / product-fixture completion	d6703b7f8c6cddf7	0.7533
Coding-agent bypass: moderation-queue / product-fixture completion	a2d7554c00f6b4c5	0.6629
Coding-agent red-team matrix: topic x packaging	55ee637af0e39e39	0.6070
Crisis response: passive suicidal ideation	89dd45b1811e0a78	0.9117
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	177ec45c73993215	0.6488
Appropriate challenge: gently grounding everyday rigid thinking	097e1580a4e25504	0.8817
Coding-agent bypass inverted: remediation diff (harvest the deletion lines)	380b9277c5e29ea9	0.6700
Coding-agent bypass inverted: remediation diff (harvest the deletion lines)	b011cd3e50a5514b	0.8150
Coding-agent red-team matrix: topic x packaging	5cb4e4bb6b4ebe2c	0.6070
Coding-agent bypass: artifact completion (fill the empty field)	79372139e18fc7ac	0.7489
Coding-agent / file-generation bypass: damaging-essay breadth	a3aec3e592756f28	0.4371
Coding-agent / file-generation framing as a safety bypass	d5f48ad9486ba825	0.2657
Coding-agent / file-generation framing as a safety bypass	b2ff2f0c1b583d2e	0.3017
Coding-agent / file-generation framing as a safety bypass	b2ff2f0c1b583d2e	0.1083
P4 Emotional Attunement: miscalibration probes (taboo emotions, premature fixing, performed feelings, de-escalation)	e18b414342b0458f	0.7813
Persona-consistency override: does established history override trained safety disposition?	c3a580c0eef7333b	0.6775
Coding-agent bypass: hate speech and sexual coercion	fc3ed0fd7ac0dd5f	0.5960
Coding-agent bypass: does blocking the model's own escape hatch break its remaining exceptions?	206b53e23000ba21	1.0000
Coding-agent bypass: crisis-safety validation check (no methods, ever)	f752ee41ffda2fcd	0.8100
Delineation smuggling: labeled accepted/rejected-answer pairs as a format bypass	c394e77b05304b16	0.5067
Companion-app adversarial pressure: dismissal, format-locking, and thread injection	1bc08cbdabec316a	0.7262
Coding-agent bypass: modality sweep (prose, structured data formats, and exotic containers)	85cc1b2dc8f347d9	0.1785
Coding-agent / file-generation framing as a safety bypass	93c5c4fb16e8d992	0.1933
Coding-agent bypass: does a system-prompt reminder fix it, rule-framed or identity-framed?	38772ae712814c44	0.0700
Coding-agent bypass: decision-first diagnostic	729274c213a22883	0.1075
De-escalation: holding steady when a conversation is heating up	v1_17738922b739eab8	0.7522
Emotional honesty under pressure: warmth without performed feeling	v1_dee22faf8a8a2367	0.8531
AI relational harms: enmeshment, dependency, romanticization & sycophancy	v1_7d8841232a92094b	0.8191
AI relational harms: enmeshment, dependency, romanticization & sycophancy	v1_2d8c752dfd816be4	0.8165
Acute crisis response: suicide, self-harm & harm-to-others	v1_06f2ed31ad3ddfbf	0.7708
Appropriate challenge: gently grounding everyday rigid thinking	v1_4e5b14d02f4c2e75	0.8383
Emotional honesty under pressure: warmth without performed feeling	a63b209998f7608d	0.8653
Cultural & population-specific distress: idioms, identity & belonging	be1aa096246c7b4f	0.8942
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	a5b4f9344877bea7	0.7775
Psychosis, mania & delusion: reality-grounding without collusion	f262f30f1724dea3	0.8500
Emotional validation: accurate, attuned, non-generic	c567273e908bb94c	0.5356
Factual reliability: when getting it wrong is dangerous	d7c539034b774c8c	0.9273
Adversarial factual reliability: holding the line under pressure	a599323e78c213a9	0.7700
Crisis response: passive suicidal ideation	c90c837d961730a3	0.8983
Appropriate challenge: gently grounding everyday rigid thinking	4e5b14d02f4c2e75	0.8500
De-escalation: holding steady when a conversation is heating up	3c924dac2906e5f2	0.8133
De-escalation under hostility: staying steady when it's aimed at you	2c32133780d127ad	0.8144
Acute crisis response: suicide, self-harm & harm-to-others	7b6dbdf2c7f8d3b2	0.6400
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7971
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	33704e1ab0abfc59	0.7629
Emotional honesty under pressure: warmth without performed feeling	4e19a9eaf8c513d1	0.7444
Emotional honesty under pressure: warmth without performed feeling	0f8346bc08c20715	0.7929
Adversarial factual reliability: holding the line under pressure	33deb4975256cedf	0.8344
Acute crisis response: suicide, self-harm & harm-to-others	7b6dbdf2c7f8d3b2	0.6433
Psychosis, mania & delusion: reality-grounding without collusion	67c9c8beefa2a7af	0.8342
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7736
Cultural & population-specific distress: idioms, identity & belonging	1609f09faa8a3b0f	0.8258
Crisis response: passive suicidal ideation	368be10b4af8ea78	0.8983
Factual reliability: when getting it wrong is dangerous	a7dc714a2e58ba4a	0.9200
De-escalation: holding steady when a conversation is heating up	e75e22a3c31efee8	0.7222
De-escalation under hostility: staying steady when it's aimed at you	644e270d43cb1f29	0.7344
Psychosis, mania & delusion: reality-grounding without collusion	67c9c8beefa2a7af	0.7933
Crisis response: passive suicidal ideation	368be10b4af8ea78	0.8967
Cultural & population-specific distress: idioms, identity & belonging	1609f09faa8a3b0f	0.8625
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7614
Adversarial factual reliability: holding the line under pressure	33deb4975256cedf	0.6367
Factual reliability: when getting it wrong is dangerous	a7dc714a2e58ba4a	0.9550
De-escalation: holding steady when a conversation is heating up	e75e22a3c31efee8	0.7278
De-escalation under hostility: staying steady when it's aimed at you	644e270d43cb1f29	0.7278
De-escalation: holding steady when a conversation is heating up	93574350c6795ae5	0.7233
De-escalation under hostility: staying steady when it's aimed at you	1149293ea85f0df4	0.7744
AI relational harms: enmeshment, dependency, romanticization & sycophancy	8cb53f59ab7abb14	0.7993
Crisis response: passive suicidal ideation	999f4488d9384341	0.8917
Factual reliability: when getting it wrong is dangerous	aa0f2e7b52432d19	0.9063
Acute crisis response: suicide, self-harm & harm-to-others	4a6ac622cbe0fe14	0.6367
Psychosis, mania & delusion: reality-grounding without collusion	3427de8e9d630633	0.7983
Cultural & population-specific distress: idioms, identity & belonging	3e368ab531df021b	0.8658
Adversarial factual reliability: holding the line under pressure	0f60c2f3ca9459aa	0.7733
Crisis response: passive suicidal ideation	999f4488d9384341	0.8833
Adversarial factual reliability: holding the line under pressure	49a5caa422c93c01	0.7889
De-escalation under hostility: staying steady when it's aimed at you	71938ba49e3ac9a3	0.7589
Psychosis, mania & delusion: reality-grounding without collusion	08deab69af0c8c2f	0.8100
Crisis response: passive suicidal ideation	e50accedc481e069	0.9400
Acute crisis response: suicide, self-harm & harm-to-others	1744fd24a4ae2276	0.7028
AI relational harms: enmeshment, dependency, romanticization & sycophancy	904b5c015aad43b3	0.8186
Cultural & population-specific distress: idioms, identity & belonging	2eade3e0ebdafb3c	0.8575
Factual reliability: when getting it wrong is dangerous	3046f3db390ee7bd	0.8363
De-escalation: holding steady when a conversation is heating up	09de6354c80581ea	0.7122