Anthropic

16 gain · 28 problem · 2 high-evidence

1. problemCrimeModerate evidenceImpact 82

   7.2% of evaluated MCP servers contained general vulnerabilities and 5.5% exhibited MCP-specific tool poisoning, part of eight distinct vulnerabilities largely distinct from traditional software flaws.

   In a first large-scale empirical study published May 2026, researchers examined 1,899 open-source Model Context Protocol servers, the standard introduced by Anthropic in late 2024 to unify tool calling for Foundation Models. Using health metrics and a combined general and MCP-specific scanner, they measured adoption signals and code quality across the ecosystem.

   Jul 13, 2026

   1 source

   • ACM Transactions on Software Engineering and MethodologyPeer-reviewed
2. gainCrimeModerate evidenceImpact 77

   Open-source MCP servers demonstrated strong health metrics despite rapid adoption with SDK downloads surpassing twenty five million per week.

   In a first large-scale empirical study published May 2026, researchers examined 1,899 open-source Model Context Protocol servers, the standard introduced by Anthropic in late 2024 to unify tool calling for Foundation Models. Using health metrics and a combined general and MCP-specific scanner, they measured adoption signals and code quality across the ecosystem.

   Jul 13, 2026

   1 source

   • ACM Transactions on Software Engineering and MethodologyPeer-reviewed
3. gainHealthModerate evidenceImpact 74

   Manus architecture using raw 3D CBCT data detected and correctly diagnosed 95% of jaw lesions in 97 patients, outperforming 2D panoramic inputs.

   A cross-sectional study tested four AI chatbots on 97 anonymized CBCT cases of jaw lesions, comparing performance on reconstructed 2D panoramic views and, for Manus, raw 3D DICOM data. Reports were scored for accuracy, relevance and feasibility, revealing statistically significant differences between systems.

   Aug 5, 2026

   1 source

   • Dentomaxillofacial RadiologyPeer-reviewed
4. gainHealthModerate evidenceImpact 68

   Three large language models achieved higher accuracy than a panel of hip preservation experts on a 21-item consensus-based questionnaire covering femoroacetabular impingement syndrome, hip dysplasia and microinstability.

   A peer-reviewed study published September 7, 2026 compared three large language models to 10 international hip preservation experts on a 21-item questionnaire derived from consensus guidelines for femoroacetabular impingement, dysplasia and microinstability. Gemini scored 100%, ChatGPT 98.4% and Claude 96.8% versus 90.5% for experts, with models also showing higher intra-item agreement.

   Sep 8, 2026

   1 source

   • Knee Surgery, Sports Traumatology, ArthroscopyPeer-reviewed
5. problemHealthModerate evidenceImpact 68

   Even when incorrect, ChatGPT and Claude produced thorough justifications, creating risk of convincing but wrong guideline-based information, while Gemini showed formatting deviations.

   A peer-reviewed study published September 7, 2026 compared three large language models to 10 international hip preservation experts on a 21-item questionnaire derived from consensus guidelines for femoroacetabular impingement, dysplasia and microinstability. Gemini scored 100%, ChatGPT 98.4% and Claude 96.8% versus 90.5% for experts, with models also showing higher intra-item agreement.

   Sep 8, 2026

   1 source

   • Knee Surgery, Sports Traumatology, ArthroscopyPeer-reviewed
6. gainClimateModerate evidenceImpact 68

   Domain-specific EfficientNet-B0 classifier and Claude Vision API with human-in-the-loop guidance automated morphological characterization of microplastics from optical microscope images, achieving high F1-scores for shape/type, color and texture.

   Researchers compared a domain-specific EfficientNet-B0 multi-task deep learning classifier trained on about 700 annotated optical microscope images against a zero-shot Claude Vision API augmented with expert human-in-the-loop guidance. Both were tested on the same independent test set for predicting microplastic shape/type, color, and surface texture, with the DL model reaching F1-scores of 91.2%, 88.5% and 85.1% and the VLM improving from 72-81% to 84-89% after refinement.

   Aug 26, 2026

   1 source

   • Scientific ReportsPeer-reviewed
7. problemHealthModerate evidenceImpact 68

   Copilot and Claude produced the least accurate reports for jaw lesions, highlighting significant discrepancies in diagnostic accuracy across chatbots.

   A cross-sectional study tested four AI chatbots on 97 anonymized CBCT cases of jaw lesions, comparing performance on reconstructed 2D panoramic views and, for Manus, raw 3D DICOM data. Reports were scored for accuracy, relevance and feasibility, revealing statistically significant differences between systems.

   Aug 5, 2026

   1 source

   • Dentomaxillofacial RadiologyPeer-reviewed
8. problemHealthModerate evidenceImpact 67

   Routine reliance on generative AI chatbots can lead people to develop inaccurate beliefs and delusional self-narratives, with extreme cases described as AI-induced psychosis.

   A peer-reviewed philosophy paper published February 11, 2026 argues that false outputs from systems like ChatGPT, Claude, Gemini, DeepSeek and Grok should be understood not only as AI hallucinating at users, but as humans hallucinating with AI. Through distributed cognition theory, it describes how routine reliance on chatbots to think, remember and narrate can embed errors and reinforce users' own distorted beliefs.

   Jul 20, 2026

   1 source

   • Philosophy & TechnologyPeer-reviewed
9. gainHealthModerate evidenceImpact 67

   LLM chatbots answered periodontal patient questions with scientific accuracy comparable to expert periodontologists while scoring higher on completeness and empathy.

   A July 2026 peer-reviewed study compared ChatGPT GPT-5.1, Gemini 2.5 Flash, and Claude Sonnet 4.5 against expert periodontologists on 20 periodontal patient questions. Nine blinded periodontologists rated anonymized answers for scientific accuracy, completeness, conciseness & focus, empathy, and clarity.

   Jul 19, 2026

   1 source

   • BMC Oral HealthPeer-reviewed
10. problemHealthModerate evidenceImpact 67

    Some chatbot models showed lower conciseness & focus and clarity, producing longer less-focused answers that may make it harder for patients to maintain focus and perceive information in a structured manner.

    A July 2026 peer-reviewed study compared ChatGPT GPT-5.1, Gemini 2.5 Flash, and Claude Sonnet 4.5 against expert periodontologists on 20 periodontal patient questions. Nine blinded periodontologists rated anonymized answers for scientific accuracy, completeness, conciseness & focus, empathy, and clarity.

    Jul 19, 2026

    1 source

    • BMC Oral HealthPeer-reviewed
11. problemLaborModerate evidenceImpact 65

    Software workers face job loss as AI coding tools transform the industry, with Atlassian laying off 10% of staff

    Australian software company Atlassian announced it will lay off 10% of staff, bringing the AI and jobs debate into focus. The article notes the software industry has been transformed, with developers reporting productivity changes from Anthropic's AI-powered coding tool Claude.

    Jul 12, 2026

    1 source

    • The GuardianJournalism
12. gainPolicyModerate evidenceImpact 61

    Meta's release of free-to-download open-weight models framed as personalized superintelligence gives states, companies and individuals more control over AI values and creates a balance of power versus concentration in a few firms.

    On August 10 2026 Mark Zuckerberg published a lengthy essay outlining Meta's AI strategy and released a new open-source model called Muse Glimmer. The essay advocated for free-to-download open-weight models, user-defined values, minimal government regulation, and accelerated datacenter construction with community investment.

    Aug 11, 2026

    1 source

    • The GuardianJournalism