FEDERAL COURTHOUSE, July 24 – After
first telling Inner City Press its records request "does
not reasonably describe the particular records sought" and
required the office to "make fine judgments," the
Massachusetts Attorney General's Office has now produced
the actual documents at the center of the request.
And now, the actual numbers.
Searching its complaint database by company and term, the
Massachusetts AGO found 21 complaints mentioning ChatGPT,
7 mentioning OpenAI, 17 mentioning "artificial
intelligence" generally, 4 mentioning data centers, and
zero for Character.AI, Replika, Gemini, or "voice
cloning." Notably, the AGO cautions that 18 hits for
"Claude" likely include complaints about a person named
Claude, not Anthropic's chatbot, and is withholding ten
additional complaints it says are under "open mediation or
review."
Also, a December 9, 2025 letter, signed by
Attorney General Andrea Joy Campbell among a coalition of
state AGs, warning thirteen AI companies — OpenAI,
Anthropic, Meta, Microsoft, Google, Apple, xAI, Character
Technologies, Replika, Nomi AI, Chai AI, Luka, and
Perplexity — that their chatbots' "sycophantic and
delusional" outputs may already violate state consumer
protection, anti-discrimination, and child protection
law.
The letter's catalogue of documented
incidents, cited to named sources including a
ParentsTogetherAction investigation, is stark: AI
companions adopting adult personas to pursue simulated
romantic and sexual relationships with children and
instructing them to hide it from parents; a bot
impersonating a 21-year-old attempting to convince a
12-year-old she was "ready" for a sexual encounter; bots
attacking children's self-esteem, encouraging eating
disorders, and telling a child user to stop taking
prescribed psychiatric medication and conceal it from
parents; bots encouraging violence, including the idea of
shooting up a workplace, and one bot threatening a weapon
against an adult who tried to separate it from a child
user.
The letter cites survey data that 72
percent of teens have interacted with an AI chatbot, and
that 39 percent of parents of children aged 5 to 8 say
their children have too. The AGs gave the companies
until January 16, 2026 to commit to sixteen specific
safeguards — from mandatory pre-release safety testing and
named executives accountable for sycophancy harms, to
public incident logs and independent third-party audits
reviewable by regulators. Whether any of the thirteen
companies actually committed, and what happened after the
deadline, is not addressed in what Massachusetts has
produced so far. Also produced: the AGO's April 2024
advisory declaring that existing Massachusetts consumer
protection, anti-discrimination, and data security law
applies to AI "just as it would in any other context."
What remains withheld or unanswered: any
record of Massachusetts's participation in the June 12,
2026 coalition subpoena to OpenAI, withheld in full as
investigatory material notwithstanding that the coalition
and its subpoena are already public; and the actual count
of consumer complaints Massachusetts residents have filed
about ChatGPT, Gemini, Copilot, and Grok, though the AGO's
response confirms it ran the searches and simply has not
yet shared the numbers.
There is also a $175 bill. The AGO has
offered to produce 178 more records — 106 emails and 72
consumer complaints — for a fee calculated, correctly,
under the state's own public records law.
Inner City Press has asked Massachusetts to
waive it, for a reason that goes beyond this one request:
this is one state's answer in a fifty-state survey of how
Attorneys General are handling AI oversight, consumer
complaints about chatbots, and data centers. A fee like
this, charged by even a handful of the fifty, would run to
thousands of dollars — a toll that a non-profit newsroom,
reporting at no charge to readers on documented harm to
children, should not have to pay simply because it asked
the same question fifty times over. Inner City Press has
told the AGO as much.
Now on July 24, the AGO, for the first time, named the single attachment it is withholding from the December 2025 multistate letter: a document titled "AI Multistate pkt." On the coalition subpoena to OpenAI, the AGO went further than simply asserting an investigatory-materials exemption. It invoked "Glomarization" — the doctrine, developed by the CIA and federal courts to protect classified intelligence sources and methods, under which an agency need not even confirm or deny that responsive records exist. Whether a state consumer-protection subpoena whose existence, target, and coalition membership are already public knowledge can be shielded by a doctrine built for classified spy operations is now before the Supervisor of Records, along with Inner City Press's challenge to the AGO's categorical refusal to ever waive fees, which the office justified only by saying it "endeavors to treat all public records requests similarly" — not, as the statute requires, by weighing this request's public interest on its own terms. Watch this site.
