Christian Chessman

intellectual property law | #appellatetwitter | kittens | archaeology | marathons | 🌈🌈🌈 | EDTX alum. | CAFC alum. | opinions mine

[Christian Chessman @cfchessman]

@SeamusHughes Taylor Farms’ safety page (¶24) reads like ChatGPT and includes unfortunate-for-them wording (“Food safety … [is] a promise we make every day”), while Taco Bell’s (¶ 29) reads like it was approved by legal (food safety is one of Taco Bell’s “priorities”).

[Christian Chessman @cfchessman]

@sebkrier Nonsense. The proposed bills granted broad, blanket immunity for AV makers, exempting them from the ordinary rules that apply to every other industry and auto-maker. The objection was to the immunity, not to incentivizing or nationally regulating AVs.

[Christian Chessman @cfchessman]

@PitBottomSuper Nonsense. The proposed bill granted broad, blanket immunity for auto-makers, exempting them from the ordinary rules that apply to every other industry and auto-maker. The objection was to the immunity, not to incentivizing or nationally regulating AVs.

[Christian Chessman @cfchessman]

@Imposter_Edits This isn’t the cause of high healthcare costs. If a patent only covers inhaler with cap, generics can legally sell inhalers without a cap. Congress already has laws letting generics re-use brands’ clinical data for free (ANDA) & gave generics a tool to challenge bad patents (IPR)

[Christian Chessman @cfchessman]

@hissgoescobra This isn’t true. South Carolina’s Supreme Court recognizes defamation for statements that “impeach the honesty or integrity or reputation … of one who is alive” *or* for those statements “tending to blacken the memory of one who is dead.” Scott v. McCain, 272 S.C. 198 (1978).

[ً @artfuIchaos]

i don't care how you were raised. you're an adult, unlearn that nonsense.

[Christian Chessman @cfchessman]

@willchen500 @JosephPatrice Rule 1.6 bars sharing client info with a 3rd party. ABA Op. 512 doesn’t bar all LLM use—it bars chats containing client data if that means the data is “disclosed to or accessed by” a 3rd party. Models that don’t train on chats still fail R1.6 *if* the AI lab can access your chats

[Christian Chessman @cfchessman]

@markankcorn @kevinbaum013 If the TOS give the AI lab zero visibility into chats (not just zero training), that seems ok. An example is OpenAI’s gpt-oss (via https://t.co/rWaoRdCjJD’s privacy interface). Encryption came up to answer your question re Google & MSFT. I agree (again) that other safeguards work

[Christian Chessman @cfchessman]

@markankcorn @kevinbaum013 "Self-learning" means the LLM trains on its own chats, in addition to its initial training data. Since LLMs can regurgitate training data, self-learning LLMs may repeat parts of chats to others. p6-7 says Al use requires privacy safeguards-encryption is one, but not the only one.

[Christian Chessman @cfchessman]

@willchen500 @JosephPatrice Oh I agree, people will get away with it. It’s still an ethics violation, though, so for lawyers who want to meet their ethics duties, it matters. It even matters for lawyers that are willing to violate ethics duties, because they may want to understand their degree of exposure.

[Christian Chessman @cfchessman]

@markankcorn @kevinbaum013 Here’s the ABA agreeing with me: Opinion 512 at p.7 says that inputting client data into generative AI models is a violation of confidentiality unless the client expressly consents or the model has encryption safeguards akin to those I described above. https://t.co/WdFyHEa9aV

[Christian Chessman @cfchessman]

@willchen500 @JosephPatrice Yes, it’s professional misconduct to disclose confident client data to others. There are *free* mainstream email & cloud storage providers with client-side encryption. I’m not sympathetic to an ethics exception for “but misconduct is convenient😢” but that’s not even true here.

[Christian Chessman @cfchessman]

@markankcorn @kevinbaum013 For that reason, Google, Box, & Microsoft offer enterprise agreements with customer-controlled encryption. They are technologically unable to view that encrypted data, akin to moving a locked box from A to B. Without similar offerings, disclosure to AI labs is…well, disclosure.

[Christian Chessman @cfchessman]

@willchen500 @JosephPatrice Cloud storage is inapt; firms use enterprise contracts with customer-controlled encryption, so the cloud provider can’t view any data it stores (eg, Microsoft’s Sovereign Cloud). Saving a file there doesn’t disclose it to Microsoft; sending it to ChatGPT discloses it to OpenAI.

[Christian Chessman @cfchessman]

@JeremiahDJohns @DoubleDawgMD Studies quantifying how people perceive probability words show that the same word often connotes radically different likelihoods. The interquartile range for “probably” ran from people thinking it meant 40% likely to people thinking it meant 95% likely. https://t.co/8tdTqofjuG

[Christian Chessman @cfchessman]

@nikicaga The US Copyright Office says purely AI-generated works aren’t copyrightable *at all,* because they lack a human author (copyright only extends to “original works of authorship” under 17 USC § 102(a)): https://t.co/IKU7rzEs4D. So they can’t be the basis of a valid copyright strike

[Christian Chessman @cfchessman]

@danepps Interestingly, the opinion is captioned “BPJ by HER next friend and mother,” tracking CA4. West Virginia’s cert and merits briefs captioned the case “BPJ by next friend and mother”—no her. BPJ’s BIO and merits briefs matched WV. Ironically, the SG’s brief says “HER next friend.”

[Christian Chessman @cfchessman]

@HabCorpLinguist @Woodxinho You think he got it down to only 83?