Source A
Anthropic rolls out Claude Opus 4.7, an AI model that is less risky than Mythos
cnbc.com
https://www.cnbc.com/2026/04/16/anthropic-claude-opus-4-7-model-mythos.html
Comparison
Winner: Source A is less manipulative
Source A appears less manipulative than Source B for this narrative.
Source B
Anthropic: Claude Opus 4.7 has a 92% honesty rate, fewer hallucinations
mashable.com
https://mashable.com/article/anthropic-claude-opus-4-7-hallucination-rate
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
Anthropic said it experimented with efforts to "differentially reduce" Claude Opus 4.7's cyber capabilities during training.
Source B main narrative
MASK honesty rate: This "tests whether a model will contradict its own stated belief when a user or system prompt pushes it to." We've already covered the MASK honesty rate, and Claude Opus 4.7 shows similar g…
Conflict summary
Stance contrast: Anthropic said it experimented with efforts to "differentially reduce" Claude Opus 4.7's cyber capabilities during training. Alternative framing: MASK honesty rate: This "tests whether a model will contradict its own stated belief when a user or system prompt pushes it to." We've already covered the MASK honesty rate, and Claude Opus 4.7 shows similar g…
Source A stance
Anthropic said it experimented with efforts to "differentially reduce" Claude Opus 4.7's cyber capabilities during training.
Stance confidence: 72%
Source B stance
MASK honesty rate: This "tests whether a model will contradict its own stated belief when a user or system prompt pushes it to." We've already covered the MASK honesty rate, and Claude Opus 4.7 shows similar g…
Stance confidence: 62%
Central stance contrast
Stance contrast: Anthropic said it experimented with efforts to "differentially reduce" Claude Opus 4.7's cyber capabilities during training. Alternative framing: MASK honesty rate: This "tests whether a model will contradict its own stated belief when a user or system prompt pushes it to." We've already covered the MASK honesty rate, and Claude Opus 4.7 shows similar g…
Why this pair fits comparison
- Candidate type: Likely contrasting perspective
- Comparison quality: 61%
- Event overlap score: 47%
- Contrast score: 71%
- Contrast strength: Strong comparison
- Stance contrast strength: High
- Event overlap: Story-level overlap is substantial. URL context points to the same episode.
- Contrast signal: Stance contrast: Anthropic said it experimented with efforts to "differentially reduce" Claude Opus 4.7's cyber capabilities during training. Alternative framing: MASK honesty rate: This "tests whether a model will cont…
Key claims and evidence
Key claims in source A
- Anthropic said it experimented with efforts to "differentially reduce" Claude Opus 4.7's cyber capabilities during training.
- Ruhani Kaur | Bloomberg | Getty ImagesAnthropic on Thursday announced a new artificial intelligence model, Claude Opus 4.7, which the company said is an improvement over past models but is "less broadly capable" than it…
- Claude Opus 4.7 is better at software engineering, following instructions, completing real-world work and is its most powerful generally available model, Anthropic said.
- But the model's cyber capabilities are not as advanced as Claude Mythos Preview, which Anthropic rolled out to a select group of companies as part of a new cybersecurity initiative called Project Glasswing earlier this…
Key claims in source B
- MASK honesty rate: This "tests whether a model will contradict its own stated belief when a user or system prompt pushes it to." We've already covered the MASK honesty rate, and Claude Opus 4.7 shows similar gains on th…
- Anthropic's reported hallucination rates are similar to the latest OpenAI models, which provide responses with incorrect information up to 5.8 percent of the time (with browsing enabled) to 10.9 percent (browsing disabl…
- Anthropic says Claude Opus 4.7 makes improvements on various types of hallucinations and overall honesty.
- Still, Claude Opus 4.7 improves upon Opus 4.6 in many ways, particularly advanced coding, visual intelligence, and document analysis, Anthropic says.
Text evidence
Evidence from source A
-
key claim
Anthropic said it experimented with efforts to "differentially reduce" Claude Opus 4.7's cyber capabilities during training.
A key claim that anchors the narrative framing.
-
key claim
Ruhani Kaur | Bloomberg | Getty ImagesAnthropic on Thursday announced a new artificial intelligence model, Claude Opus 4.7, which the company said is an improvement over past models but is…
A key claim that anchors the narrative framing.
-
evaluative label
What we learn from the real-world deployment of these safeguards will help us work towards our eventual goal of a broad release of Mythos-class models." Since its founding in 2021, Anthropi…
Evaluative labeling that nudges a normative interpretation.
-
selective emphasis
Claude Opus 4.7 is available across all of Anthropic's Claude products, its application programming interface and through cloud providers Microsoft, Google and Amazon.
Possible selective emphasis on specific aspects of the story.
Evidence from source B
-
key claim
Anthropic says Claude Opus 4.7 makes improvements on various types of hallucinations and overall honesty.
A key claim that anchors the narrative framing.
-
key claim
Still, Claude Opus 4.7 improves upon Opus 4.6 in many ways, particularly advanced coding, visual intelligence, and document analysis, Anthropic says.
A key claim that anchors the narrative framing.
-
evaluative label
More details on Claude Opus 4.7 hallucination ratesWhen using Opus 4.7, how likely is Claude to tell a lie, invent facts, or deceive users?
Evaluative labeling that nudges a normative interpretation.
-
causal claim
There isn't a single hallucination rate that Anthropic provides, because there are multiple types of hallucinations.
Cause-effect claim shaping how events are explained.
-
selective emphasis
This shows just how stubborn AI hallucinations are, with even leading AI companies like Anthropic recording input hallucination rates around 90 percent.
Possible selective emphasis on specific aspects of the story.
Bias/manipulation evidence
-
Source A · Framing effect
Claude Opus 4.7 is available across all of Anthropic's Claude products, its application programming interface and through cloud providers Microsoft, Google and Amazon.
Possible framing pattern: wording sets a specific interpretation frame rather than neutral description.
-
Source B · Appeal to fear
This shows just how stubborn AI hallucinations are, with even leading AI companies like Anthropic recording input hallucination rates around 90 percent.
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
27%
emotionality: 29 · one-sidedness: 30
Detected in Source A
framing effect
Source B
39%
emotionality: 41 · one-sidedness: 35
Detected in Source B
appeal to fear
Metrics
Framing differences
- Source A emotionality: 29/100 vs Source B: 41/100
- Source A one-sidedness: 30/100 vs Source B: 35/100
- Stance contrast: Anthropic said it experimented with efforts to "differentially reduce" Claude Opus 4.7's cyber capabilities during training. Alternative framing: MASK honesty rate: This "tests whether a model will contradict its own stated belief when a user or system prompt pushes it to." We've already covered the MASK honesty rate, and Claude Opus 4.7 shows similar g…
Possible omitted/downplayed context
- Review which economic and policy factors each source keeps outside focus.
- Check whether alternative explanations are acknowledged.