Data Report · RedactSure Research
Chris Sowa · Founder, RedactSure
Published
Last updated
How Did Developers React to Meta Muse and OpenAI Dots?
Of 741 developer comments on Meta Muse, 109 raised security or privacy concerns, about 1 in 7. Of 646 developer comments on OpenAI Dots, 101 did, about 1 in 6. Business-buyer and consumer opinion could not be collected.
About one in seven developer comments on Meta Muse raised a security or privacy concern: 109 of 741. On OpenAI Dots the share was about one in six, 101 of 646. The counts come from RedactSure's coding of developer comments about each product, and they describe what developers chose to raise, not whether either product was found unsafe.
Key findings
- Meta Muse: 109 of 741 developer comments raised security or privacy concerns, 14.7 percent, about 1 in 7 (RedactSure Research, 2026).
- Meta Muse: privacy was raised in 86 of the 741 developer comments, about 1 in 9, and these are part of the 109 that raised security or privacy (RedactSure Research, 2026).
- OpenAI Dots: 101 of 646 developer comments raised security or privacy concerns, 15.6 percent, about 1 in 6 (RedactSure Research, 2026).
How do the two products compare?
| Product | Developer comments | Raised security or privacy | Rate |
|---|---|---|---|
| Meta Muse | 741 | 109 | 14.7%, about 1 in 7 |
| OpenAI Dots | 646 | 101 | 15.6%, about 1 in 6 |
| Meta Muse, privacy (part of the 109) | 741 | 86 | about 1 in 9 |
The two shares sit in a similar range. This report does not place them beside coding agents or other products: the products with higher shares also differ in brand and timing, and these counts cannot separate those causes from the product itself.
What did developers raise?
The measure counts a comment once when it raised security or privacy, whether it named one or both. The Meta Muse privacy figure counts every comment that raised privacy, so those 86 sit inside the 109 rather than adding to them. A comment was read in full and coded once, so a long thread does not add weight through repetition.
These are counts of comments, not of people. One person may comment more than once, and a single comment may raise several topics. The counts say which concerns appeared, not how strongly each commenter held them.
What do the numbers mean for teams running AI agents?
A developer comment is not a control test. It shows that the topic came up when the product was discussed, not that the product failed in use. The control question is separate and more concrete: what reaches the model when it reads a screen, who approves an action before it completes, and what record exists afterward.
That is where a team can act. A count of concerns does not tell you what the agent saw on a particular invoice or account page, and it does not tell you whether a person approved the payment. Those answers come from the workflow, not from the comment thread.
Methodology and limitations
RedactSure read and coded developer comments about AI agent products: Hacker News threads for each product, plus GitHub issues for Claude Code. Every comment was read and coded for sentiment and for six concern tags, one of which is security or privacy. Counts are aggregated; no comment text or commenter is quoted, named or linked.
The comment counts below name every product reviewed for this report. The collection window runs from each product's release to 30 days after it. A rate is shown only where a product has 30 or more comments.
| Product | Source | Developer comments | Collection window | Rate |
|---|---|---|---|---|
| OpenAI Dots | Hacker News | 646 | September 29 to October 29, 2026 | 15.6% |
| Meta Muse | Hacker News | 741 | September 8 to October 8, 2026 | 14.7% |
| Claude Code | Hacker News and GitHub | 1,530 | February 24 to March 26, 2025 | 2.6% |
| Claude Computer Use | Hacker News | 90 | October 22 to November 21, 2024 | 3.3% |
| Codex | Hacker News | 134 | April 16 to May 16, 2025 | 3.0% |
| Devin | Hacker News | 542 | March 12 to April 11, 2024 | 0.4% |
| Comet | Hacker News | 19 | July 9 to August 8, 2025 | Count only |
| Manus | Hacker News | 19 | March 13 to April 12, 2025 | Count only |
Business-buyer and consumer opinion could not be collected. Review sites are invited and skew positive, and app-store reviews of a host app do not name the agent inside it. Neither audience is represented in the figures above.
The limits that travel with the figures: these are developer comments, not buyers or the public; they were collected in the weeks after each product's release, not in steady state; one coder read every item; the counts are aggregated and no comment is quoted; and a high rate means the topic was raised, not that the product was found unsafe. No audience comparison and no cross-product comparison is made. The underlying data and the coding method are held by RedactSure Research and are described in the data set cited below.
Frequently asked questions
What does 14.7 percent mean in this report?
It is the share of 741 developer comments about Meta Muse that raised security or privacy: 109 comments. The comment was counted once whether it named one concern or both, and the count is of comments, not of people.
Are these developer comments the same as product reviews?
No. They are comments in developer discussion threads, not reviews, and they are not a sample of buyers or the public. A comment shows that someone raised a topic, not that a product failed or that a buyer would agree.
Can I compare Meta Muse and OpenAI Dots with coding agents?
This report does not make that comparison. The products with higher shares also differ in brand and timing, and the counts cannot separate those causes from the product type. Treat the per-product figures as descriptions of each discussion, not as a ranking of risk.
What should a team take from the number?
Use it as a prompt to ask the control questions: what the model reads on a screen, who approves an action before it completes, and what record exists after it. The comment count cannot answer any of those for a specific workflow.
How RedactSure Closes the Gap Between a Comment and a Control
A developer comment can name a concern, and it cannot show what the model read on the screen behind it. RedactSure's environment turns the screen into tokens before the model sees it, so the account number or the patient name is not in the context, and a named person approves each consequential action.
The environment that enforces this is built by RedactSure, an AI agent controls, governance and data protection company. Before, the agent reads the page and acts; after, the model receives tokens, the person approves, and the record shows both. Enterprise deployments are in pilot.
Everyone else controls where data can go. RedactSure controls what the AI can see.
See it on your own workflow
Bring one agent workflow and see, with test values, what the model reads and who approves each action. The count tells you the topic came up; the workflow tells you what your agent actually sees. Book a demo