Try OraCore Scribe free for 14 days. No credit card required.
Welcome to this week’s Dental AI Weekly, honest analysis of where dental AI is going, from someone building in it.
OpenAI introduced software that can keep working in the background through connected apps. Two senators proposed a bill that would assign civil and criminal liability when AI agents are used in hacking. Google DeepMind tested a detectable signature inside AI-generated protein designs.
The common thread is the evidence AI leaves behind.
A useful AI system will still be judged by its results. As that software keeps working beyond a single request, the record around those results will matter too.
WHAT HAPPENED THIS WEEK
01 – BACKGROUND AGENTS
OpenAI introduced dots, agents that can continue working through connected workplace apps instead of waiting for one request at a time.
OpenAI says each dot has its own cloud computer, can connect to apps, appear in Slack or Teams, follow custom rules, and perform background research. The product includes approval controls and an activity view that records what the software has done.
The September 29 announcement is a product launch, not an independent performance study. Availability depends on plan, market, and workspace settings.
Read OpenAI’s September 29 announcement
What this could mean for dentistry, OraCore editorial analysis: The important change is persistence. Most dental AI waits for someone to start a task. Software built on this pattern could keep preparing work across the day while the team treats patients. When AI keeps working after the prompt, the history it leaves behind becomes part of the product. It gives the practice a way to reconstruct what the software prepared, changed, or sent.
02 – ACCOUNTABLE ACTIONS
Senators Josh Hawley and Chris Murphy introduced a bill aimed at developers and operators when AI agents are used to commit computer crimes.
Their October 1 announcement says the AI Agent Accountability Act would apply civil and criminal liability under the Computer Fraud and Abuse Act and allow enforcement by federal and state attorneys general. The proposal responds to agents that can take several steps toward a goal rather than simply return an answer.
This is introduced legislation, not law. The sponsor announcement establishes what the bill proposes, not whether it will pass or how courts would apply it.
Read the October 1 Senate announcement
What this could mean for dentistry, OraCore editorial analysis: The proposal does not set rules for dental software. It does show lawmakers beginning to treat an agent’s harmful action as conduct with accountable people behind it, rather than as a mysterious software failure. That idea will matter more as connected software takes on real work inside healthcare businesses.
03 – A SIGNATURE INSIDE THE DESIGN
Google DeepMind introduced SynthID Bio, an early method for placing a detectable signature inside AI-generated protein sequences and predicted structures.
Google DeepMind says it tested the method across three target proteins and found that the signature could be added without preventing the tested designs from functioning in initial laboratory work. The company published its methods, code, data, and model weights.
SynthID Bio is a proof of concept, not a clinical product. A detectable origin does not prove that a design is safe or effective, and the method still has to withstand attempts to remove or defeat the signature.
Read Google DeepMind’s September 30 announcement
What this could mean for dentistry, OraCore editorial analysis: Dentistry may eventually use AI-designed biological components in salivary diagnostics, antimicrobial treatments, regenerative materials, or other oral-health research. Researchers and regulators may then need to distinguish a sequence designed by AI from one found in nature or developed through a conventional process. SynthID Bio is an early attempt to make origin detectable inside the design itself.
“When AI keeps working after the prompt, the history it leaves behind becomes part of the product.”
THE BIGGER SIGNAL
AI has spent the last few years getting better at producing an answer. These developments point to the next problem: preserving evidence around the work.
OpenAI is making an activity history part of a background agent. Senators are proposing that responsibility follow the people behind harmful agent actions. Google DeepMind is testing whether an AI-generated biological design can carry a detectable record of origin inside the design itself.
For dentistry, this reaches beyond compliance. A clinical note, a prepared patient message, a schedule change, or a future AI-designed biomaterial will each need a trustworthy history appropriate to the work. The final result matters. So does the trail that explains where it came from.
FROM THE ORACORE BLOG THIS WEEK
Dental AI | 3 min
Connected AI can do more useful work when it understands the systems and context already inside a dental practice. This article looks at what changes when the software is built around the practice management system rather than beside it.
AI OPTIMISM
The encouraging part is that the most serious AI builders are starting to design for memory, responsibility, and provenance.
That matters for dental teams because the useful version of AI is not a black box that asks the practice to trust it blindly. It is software that helps with real work and leaves enough context behind for the owner, provider, or team to understand what happened.
BY THE NUMBERS
233
reported AI-related incidents were recorded for 2024 by the AI Incidents Database, a record high and a 56.4% increase from 2023.
What it signals: More documented incidents make the record around an AI system’s actions harder to treat as a technical detail. The count covers reported incidents across industries, not dentistry, and does not capture every incident that occurred.
Source: Stanford Institute for Human-Centered Artificial Intelligence, 2025 AI Index, April 7, 2025.
READER Q&A
“Does an AI scribe create the clinical note, or will our team still have to turn the recording into one?”
BH: A true dental AI scribe should create the clinical note for you.
I have seen providers use tools like Plaud, then email the transcript to themselves, run it through ChatGPT, reformat the result so it copies correctly, and finally paste it into the PMS. That is not a convenient workflow. It also creates real HIPAA concerns when patient conversations move between email, consumer AI tools, and accounts that may not have the proper safeguards or agreements in place.
OraCore handles that work inside one dental workflow. It listens to the appointment and creates an appointment-specific clinical note with the tooth numbers, procedures, findings, materials, and next steps already organized. The transcript is available as supporting context, but your team does not have to turn it into the note.
The appointment should end with the clinical note ready, not with five more steps for your team.
See how OraCore Scribe turns a dental appointment into a complete, appointment-specific clinical note. Scribe Solo and Scribe Team include a free 14-day trial with no credit card required.
That’s the week. Reply with what you’re seeing in your own practice. I read every one.
Brad Hutchison
CEO, OraCore AI
Dental AI Weekly drops every Monday. pulse@oracorenews.com | OraCore AI, Denver CO