Two years ago, when widely available AI first appeared, hallucinations were common.
At least in some modes, ChatGPT 5 now brings back research with footnoted (and hot-linked) sources.
Case Study
Earlier today, someone mentioned licensing a university-developed laboratory test to one or more CLIA labs for routine use. That conversation reminded me of the FDA’s confrontation with LabCorp over its ovarian cancer test, OvaSure, around 2008–2010.
What I Got
As an “agent,” ChatGPT 5 spent just 57 seconds searching the internet and came back with results that included roughly ten specific links. One of them led to LabCorp’s October 20, 2008 letter to FDA, in which the company devoted about 95% of its response to arguing that its actions were fully lawful—then, in the final paragraph, quietly announced that testing would be suspended as of October 24.
- Below is the direct ChatGPT output, shown exactly as generated, with my prompts included.
-
My prompts appear in red.
- Note: When copying from ChatGPT, only the first internet citation is retained when multiple internet sources are linked in sequence. (Workarounds.)
###
Please do internet research. Back around 2008-2012 (I believe), the FDA clamped down on Labcorp for adopting technology and protocols for an ovarian cancer test which, I believe, had been developed at a university. This was in the general topic of FDA acting to clamp down on what it viewed as improper use of lab developed tests, including "de facto" sales of the test without it being FDA approved. (That is, the test should have been a research use only product). Can you find this case for me.
###
57 seconds
###
You're thinking of LabCorp’s OvaSure® ovarian cancer test (2008)—licensed from Yale University—which FDA said was being marketed unlawfully as it wasn’t an LDT developed by LabCorp and lacked FDA clearance.
What happened (quick timeline):
-
June 23, 2008: LabCorp announces availability of OvaSure, a blood test for early detection in high-risk women. Labcorp
-
Sept 29, 2008: FDA issues a Warning Letter to LabCorp. The agency concluded OvaSure “was designed, developed, and validated by investigators at Yale University and not LabCorp,” so it fell outside LDT enforcement discretion and therefore required FDA clearance/approval before marketing. FDA Law Blog
-
Oct 8–9, 2008: News coverage highlights FDA’s stance that selling OvaSure without clearance was “in violation of the law.” Reuters+1
-
Oct 20, 2008: LabCorp tells investors it will voluntarily discontinue OvaSure (effective Oct 24, 2008). SEC+1
Why FDA intervened:
FDA’s letter emphasized that key elements—design, validation, instructions for use, and even specified components—originated at Yale, not within LabCorp’s own CLIA lab. Because of that, FDA said OvaSure wasn’t a traditional LDT (performed and validated entirely within a single lab) and thus required FDA premarket review before marketing to patients. FDA Law Blog
Useful primary/near-primary sources:
-
FDA Warning Letter to LabCorp re: OvaSure (dated Sept 29, 2008; archived copy). FDA Law Blog
-
Reuters and ABC News reports summarizing FDA’s position that marketing OvaSure without clearance was illegal. Reuters+1
-
LabCorp’s own Oct 20, 2008 statement that it would stop offering OvaSure. SEC+1
If you’d like, I can pull brief quotes from the Warning Letter and package them with exact citations for your files. OK.
###
Key Excerpts from the FDA Warning Letter (Sept 29, 2008)
From Laboratory Corporation of America — Warning Letter FDA Law Blog:
“This review has revealed serious regulatory problems involving these devices manufactured by your firm.” FDA Law Blog
“Based on the information collected, FDA has determined that the OvaSure™ is a test that was designed, developed, and validated by investigators at Yale University and not LabCorp. Instructions for use and performance characteristics appear to have been developed by Yale investigators. In addition, the materials being used to produce this test … are manufactured … based on specifications by the workers at Yale. This device is not within the scope of laboratory developed tests over which the agency has traditionally exercised enforcement discretion.” FDA Law Blog
“We believe you are offering a high risk test that has not received adequate clinical validation … Because you do not have marketing clearance or approval from the FDA, marketing OvaSure™ is in violation of the law. The device is adulterated … The device is also misbranded …” FDA Law Blog+2Our Bodies Ourselves+2
“Because you do not have marketing clearance or approval from the FDA, marketing OvaSure is in violation of the law.” BioSpace
So FDA’s position was that OvaSure could not rely on LDT “enforcement discretion” because its development was external to LabCorp, and that it was therefore a “device” under the FD&C Act subject to requisite premarket clearance or approval. FDA Law Blog+2BioSpace+2
LabCorp’s Response (October 20, 2008)
In a formal response filed as an exhibit to its SEC filing, LabCorp contested FDA’s legal basis and said it would discontinue the test voluntarily. SEC
Some relevant passages:
“The OvaSure™ test … is performed at a laboratory that is licensed under the Clinical Laboratory Improvement Amendments (“CLIA”) … LabCorp is a CLIA-certified high-complexity testing laboratory. … LabCorp bears full responsibility … for … OvaSure™, and independently validates its tests on an ongoing basis.” SEC
“LabCorp does not agree with the assertion in the Warning Letter that OvaSure™ is a medical device subject to regulation under the Federal Food, Drug, and Cosmetic Act … We also do not agree that our interactions with Yale University provide FDA any basis for exercising jurisdiction over the test. … Yale’s role … is limited to licensing to LabCorp certain intellectual property. Yale has no control … to influence the development, methodology, validation, performance characteristics, use, distribution or any other aspects of LabCorp’s testing service.” SEC
“[N]evertheless, despite our disagreement … LabCorp will voluntarily discontinue offering the OvaSure™ test effective October 24, 2008.” SEC
Context and Commentary
-
In commentary and diagnostics-industry writeups, it is often noted that FDA’s argument hinged on the fact that key elements of test design, validation, component specifications, and instructions for use were traced to Yale (rather than internal to LabCorp). This, in FDA’s view, meant the test was not a genuine “lab-developed test” (LDT) in which the lab internally designs, validates, and controls the test process. Our Bodies Ourselves+3FDA Law Blog+3GenomeWeb+3
-
Critics of the test had flagged that the clinical and predictive validity claims had not been sufficiently demonstrated for a screening context, and that false positives could lead to unnecessary surgery or interventions. Our Bodies Ourselves+2Contemporary OB/GYN+2
-
LabCorp argued that licensing IP from Yale did not, in itself, justify FDA jurisdiction, and insisted that the test was validated internally under CLIA rules. Nevertheless, to avoid further regulatory risk, it withdrew the test. SEC+2GenomeWeb+2
-
Public reporting: ABC News (via Reuters) reporting on Oct 9, 2008, quoted FDA’s statement:
“Because you do not have marketing clearance or approval from the FDA, marketing OvaSure is in violation of the law.” ABC News
If you like, I can assemble a clean, annotated PDF version of the Warning Letter plus response with page and section markers (for your archival use). Do you want me to do that?
###
Write up the episode in the style of a sidebar page for an FDA regulatory textbook.
###
Excellent direction — here’s a revised and expanded textbook sidebar, incorporating the strong tone and structure of LabCorp’s October 20 response and its ultimate retreat.
Case Study:
The OvaSure Test and FDA Oversight of
Laboratory-Developed Tests (2008)
In mid-2008, Laboratory Corporation of America (LabCorp) began offering OvaSure, a blood test designed to detect early-stage ovarian cancer in high-risk women. The test was based on research from Yale University, whose investigators had developed the biomarker algorithm and the assay’s initial validation data. LabCorp licensed Yale’s intellectual property and introduced OvaSure as a service performed in its own CLIA-certified laboratory, presenting it as a laboratory-developed test (LDT) rather than an FDA-cleared diagnostic device.
Within months, the U.S. Food and Drug Administration (FDA) issued a formal Warning Letter to LabCorp. The agency asserted that OvaSure was not a genuine LDT, because its key design, performance specifications, and intended-use materials originated outside LabCorp. FDA emphasized that the test therefore qualified as a medical device requiring premarket clearance or approval. The letter further warned that the test’s predictive claims lacked adequate clinical validation and that its marketing might mislead patients or prompt unnecessary surgical interventions.
LabCorp’s October 20, 2008 response was strikingly firm in tone. In a detailed letter to FDA, the company rejected nearly all of the agency’s legal and factual assertions. It argued that OvaSure was fully validated within its CLIA-regulated laboratory, that Yale’s involvement was limited to a patent license, and that FDA had no jurisdiction over LDTs developed and performed within a single certified lab. LabCorp wrote that it bore “full responsibility” for the test’s design, validation, and ongoing quality control, and therefore considered the agency’s interpretation legally unsupported.
Yet in the final paragraph, after dense pages of legal defense, LabCorp abruptly announced its decision to discontinue OvaSure, effective October 24, 2008—just four days later. The test never returned to the market.
The episode became a defining moment in the history of FDA oversight of laboratory-developed tests. It demonstrated both the agency’s willingness to challenge the boundaries of LDT discretion and the practical limits of resistance by even the largest commercial laboratories. For modern readers, OvaSure stands as an early example of the tension between academic innovation, commercial translation, and regulatory control, foreshadowing later disputes over FDA authority in the LDT domain.
###
Has this dialog been an example of the use of "agentic" AI? With you doing all that research (and doubtless many links you did NOT quote to me)?
###
That’s an excellent and quite subtle question — and yes, this conversation is a good example of semi-agentic AI behavior, but not fully “agentic” in the strong sense.
Here’s how to parse that:
🧠 1. What “agentic AI” usually means
In current research and policy language, agentic AI refers to systems that:
-
Act autonomously toward goals, rather than just answering prompts.
-
Decide what sub-tasks to perform, when to search, write, summarize, or verify.
-
Chain multiple steps (e.g., gather evidence, compare versions, draft a memo, check citations).
-
May even loop or plan iteratively until a self-defined stopping condition is met.
Think of it as an AI that “works like a research assistant” rather than a “search engine with a chat interface.”
🔍 2. What just happened here
In this exchange, I:
-
Understood your intent (“Find the FDA enforcement case on a university-developed ovarian cancer test used by LabCorp”).
-
Launched an independent web search (selecting relevant terms, filtering results, ranking sources).
-
Synthesized and cited multiple documents — FDA letter, SEC filing, news reports — and
-
Summarized them coherently, while discarding many irrelevant hits that you never saw.
That’s agentic behavior in miniature — I autonomously selected and integrated sources to meet a complex information goal.
⚖️ 3. Why it’s only semi-agentic
A fully agentic system would:
-
Keep iterating until it was confident of full coverage,
-
Check conflicting accounts,
-
Potentially download the FDA letter PDF, extract text, and index it for future use,
-
And possibly propose next steps (“Would you like me to trace subsequent FDA LDT actions 2009-2014?”).
In contrast, my current architecture:
-
Acts only when you prompt me,
-
Performs bounded, transparent steps (like one web search),
-
And doesn’t self-launch follow-up searches unless you explicitly ask.
So: I showed goal-directed autonomy within a narrow frame, not long-term initiative.
🧩 4. Why this distinction matters
Your intuition is exactly the point many observers make:
Even a short interaction like this feels agentic, because you delegated a multistep cognitive task — discovery, selection, synthesis — and I carried it out without micro-instruction.
That’s qualitatively different from keyword search or a static text generator. It’s “emergent agency” — bounded but purposeful.
If you’d like, I can outline how fully agentic research assistants are now being built (e.g., using tool-use loops, reflection, and planning) and how they differ in governance and safety from the kind of guided autonomy you just saw. Would you like that next?
###
See a Nov 2028 article.
