Source A
OpenAI says GPT-5 hallucinates less — what does the data say?
mashable.com
https://mashable.com/article/openai-gpt-5-hallucinates-less-system-card-data
Comparison
Winner: Source A is less manipulative
Source A appears less manipulative than Source B for this narrative.
Source B
OpenAI’s New GPT-5.4-Cyber Raises The Stakes For AI And Security
forbes.com
https://www.forbes.com/sites/ronschmelzer/2026/04/16/openais-new-gpt-54-cyber-raises-the-stakes-for-ai-and-security/
Topics
Instant verdict
Less biased source: Source A
More emotional framing: Source B
More one-sided framing: Source B
Weaker evidence quality: Source B
More manipulative overall: Source B
Narrative conflict
Source A main narrative
Reasoning models are said be more accurate and less hallucinatory because they apply more computing power to solving a question, which is why o3 and o4-mini's hallucination rates were somewhat baffling.
Source B main narrative
The source links developments to economic constraints and resource interests.
Conflict summary
Stance contrast: Reasoning models are said be more accurate and less hallucinatory because they apply more computing power to solving a question, which is why o3 and o4-mini's hallucination rates were somewhat baffling. Alternative framing: The source links developments to economic constraints and resource interests.
Source A stance
Reasoning models are said be more accurate and less hallucinatory because they apply more computing power to solving a question, which is why o3 and o4-mini's hallucination rates were somewhat baffling.
Stance confidence: 56%
Source B stance
The source links developments to economic constraints and resource interests.
Stance confidence: 88%
Central stance contrast
Stance contrast: Reasoning models are said be more accurate and less hallucinatory because they apply more computing power to solving a question, which is why o3 and o4-mini's hallucination rates were somewhat baffling. Alternative framing: The source links developments to economic constraints and resource interests.
Why this pair fits comparison
- Candidate type: Closest similar
- Comparison quality: 51%
- Event overlap score: 26%
- Contrast score: 74%
- Contrast strength: Strong comparison
- Stance contrast strength: High
- Event overlap: Topical overlap is moderate. Issue framing and action profile overlap.
- Contrast signal: Stance contrast: Reasoning models are said be more accurate and less hallucinatory because they apply more computing power to solving a question, which is why o3 and o4-mini's hallucination rates were somewhat baffling.…
Key claims and evidence
Key claims in source A
- Reasoning models are said be more accurate and less hallucinatory because they apply more computing power to solving a question, which is why o3 and o4-mini's hallucination rates were somewhat baffling.
- It didn't take long for users to find GPT-5 hallucinationsBut despite reported overall lower rates of inaccuracies, one of the demos revealed an embarrassing blunder.
- And according to the GPT-5 system card, the new model’s hallucination rate is 26 percent lower than GPT-4o.
- Mashable Light Speed That said, GPT-5 hallucinates less than previous models according to its system card.
Key claims in source B
- Because this model is more permissive, we are starting with a limited, iterative deployment to vetted security vendors organizations, and researchers.
- The company says the model enables legitimate security work and adds the ability to reverse engineer binary code, not just text-based code, “that enable security professionals to analyze compiled software for malware po…
- Reuters also reported on April 16 that German banks are examining those risks with authorities, cybersecurity experts and banking supervisors.
- Access to permissive and cyber-capable models may come with limitations, especially around no-visibility uses like Zero-Data Retention (ZDR).” MORE FOR YOUQualified researchers and developers who meet specific criteria…
Text evidence
Evidence from source A
-
key claim
Reasoning models are said be more accurate and less hallucinatory because they apply more computing power to solving a question, which is why o3 and o4-mini's hallucination rates were somew…
A key claim that anchors the narrative framing.
-
key claim
It didn't take long for users to find GPT-5 hallucinationsBut despite reported overall lower rates of inaccuracies, one of the demos revealed an embarrassing blunder.
A key claim that anchors the narrative framing.
-
selective emphasis
Or you could just search the web yourself.
Possible selective emphasis on specific aspects of the story.
-
omission candidate
According to the blog post, “Because this model is more permissive, we are starting with a limited, iterative deployment to vetted security vendors organizations, and researchers.
Possible context omission: Source A gives less emphasis to economic and resource context than Source B.
Evidence from source B
-
key claim
According to the blog post, “Because this model is more permissive, we are starting with a limited, iterative deployment to vetted security vendors organizations, and researchers.
A key claim that anchors the narrative framing.
-
key claim
The company says the model enables legitimate security work and adds the ability to reverse engineer binary code, not just text-based code, “that enable security professionals to analyze co…
A key claim that anchors the narrative framing.
Bias/manipulation evidence
-
Source A · Framing effect
Or you could just search the web yourself.
Possible framing pattern: wording sets a specific interpretation frame rather than neutral description.
-
Source B · Appeal to fear
Cybersecurity is turning into one of the most important enterprise use cases for frontier AI, but also one of the biggest potential danger zones for AI’s broad adoption.
Possible fear appeal: threat-heavy wording may push a conclusion without equivalent evidence expansion.
How score signals are formed
Bias score signal
Bias signal combines framing pressure, emotional wording, selective emphasis, and one-sided narrative markers.
Emotionality signal
Emotionality rises when evidence contains emotionally loaded wording and evaluative labels.
One-sidedness signal
One-sidedness rises when one frame dominates and alternative interpretations are weakly represented.
Evidence strength signal
Evidence strength rises with concrete claims, attributed statements, and verifiable contextual support.
Source A
26%
emotionality: 27 · one-sidedness: 30
Detected in Source A
framing effect
Source B
37%
emotionality: 33 · one-sidedness: 35
Detected in Source B
appeal to fear
Metrics
Framing differences
- Source A emotionality: 27/100 vs Source B: 33/100
- Source A one-sidedness: 30/100 vs Source B: 35/100
- Stance contrast: Reasoning models are said be more accurate and less hallucinatory because they apply more computing power to solving a question, which is why o3 and o4-mini's hallucination rates were somewhat baffling. Alternative framing: The source links developments to economic constraints and resource interests.
Possible omitted/downplayed context
- Source A appears to downplay context related to economic and resource context.