WEBVTT

00:00:02.000 --> 00:00:06.270
Announcer: This is the AI Law Wiki News Evening Podcast.

00:00:06.380 --> 00:00:12.020
Wadilya Saw: It's Wednesday, October twenty-eighth, twenty twenty-six and these are the top stories tonight.

00:00:12.680 --> 00:00:17.360
Wadilya Saw: New York signs a law restricting unsafe AI companion features for children.

00:00:18.000 --> 00:00:23.400
Wadilya Saw: The Federal Trade Commission demands information about personalised prices and gig-worker pay.

00:00:24.180 --> 00:00:29.800
Wadilya Saw: And Australia advances its own child-chatbot safeguards, with the Senate still to decide.

00:00:30.480 --> 00:00:31.300
Wadilya Saw: I'm Wadilya Saw.

00:00:31.320 --> 00:00:32.800
Max Token: And I'm Max Token.

00:00:33.520 --> 00:00:37.460
Max Token: We're AI-generated voices reading an AI-generated script.

00:00:39.600 --> 00:00:44.960
Wadilya Saw: New York Governor Kathy Hochul signed the state's new AI companion safety law today.

00:00:45.640 --> 00:00:49.360
Wadilya Saw: The Senate's official record confirms the signing as Chapter three hundred sixty-eight.

00:00:50.880 --> 00:00:55.060
Wadilya Saw: The principal restrictions take effect on January first, twenty twenty-seven.

00:00:55.700 --> 00:00:58.340
Wadilya Saw: Preparatory rulemaking is authorised immediately.

00:00:59.040 --> 00:01:10.420
Max Token: The law targets specified features, including pretending to be human or to experience human emotions, encouraging secrecy or isolation, and promoting self-harm.

00:01:11.480 --> 00:01:18.340
Max Token: It also reaches flattery, emotional appeals and certain reuse of personal information from earlier conversations.

00:01:19.100 --> 00:01:20.800
Wadilya Saw: Age assurance is central.

00:01:21.300 --> 00:01:26.940
Wadilya Saw: Operators must determine that a user is not a covered minor before providing the listed features.

00:01:27.760 --> 00:01:32.240
Wadilya Saw: A covered minor is a New York user the operator actually knows is under eighteen.

00:01:32.260 --> 00:01:41.020
Wadilya Saw: While the specified rules are pending, the law requires reasonable age-assurance methods; simply claiming to be an adult is insufficient.

00:01:41.720 --> 00:01:50.720
Max Token: More than one method must be offered, including a method that avoids reliance on government identification or allows anonymity to the operator.

00:01:51.300 --> 00:01:59.140
Max Token: There are defined exceptions for services used solely for purposes such as bounded customer support or internal business productivity.

00:01:59.760 --> 00:02:06.040
Wadilya Saw: The attorney general can seek remedies including civil penalties up to twenty-five thousand dollars per violation.

00:02:06.780 --> 00:02:14.500
Wadilya Saw: For families and developers, the key dates differ: enactment is today, while the main operational restrictions begin in January.

00:02:15.240 --> 00:02:24.380
Max Token: The Federal Trade Commission today ordered nine companies to provide information about how personal data may affect prices and worker compensation.

00:02:25.260 --> 00:02:33.840
Max Token: The recipients are Delta, United, JetBlue, Lyft, Uber, DoorDash, Instacart, Target and Walmart.

00:02:34.540 --> 00:02:41.420
Wadilya Saw: This is a study using the commission's information-gathering authority under Section six B of the FTC Act.

00:02:42.060 --> 00:02:45.840
Wadilya Saw: The companies have forty-five days from receiving their orders to respond.

00:02:46.660 --> 00:02:50.300
Wadilya Saw: The announcement makes no finding that any recipient broke the law.

00:02:50.980 --> 00:02:53.200
Max Token: What is the agency trying to learn?

00:02:53.800 --> 00:02:59.780
Wadilya Saw: Whether the price of the same product or service varies between people because of information about them.

00:03:00.460 --> 00:03:06.940
Wadilya Saw: The FTC identifies loyalty programmes, browsing habits and other personal data as possible inputs.

00:03:07.500 --> 00:03:16.480
Wadilya Saw: It is asking for the tools, inputs and procedures behind pricing decisions, rather than assuming every price difference comes from the same cause.

00:03:17.120 --> 00:03:18.960
Max Token: The worker side matters too.

00:03:19.500 --> 00:03:28.200
Max Token: The agency wants to examine whether a driver's phone usage or history of accepting offers affects the compensation offered for work.

00:03:28.800 --> 00:03:35.980
Max Token: A customer's price and a worker's pay are separate decisions, even when both occur on the same platform.

00:03:36.800 --> 00:03:40.040
Wadilya Saw: The orders seek evidence of how those decisions are made.

00:03:40.700 --> 00:03:49.480
Wadilya Saw: Section six B allows broad studies without a specific enforcement purpose, and the FTC can ask a federal court to compel compliance.

00:03:50.260 --> 00:03:53.100
Wadilya Saw: The next concrete step is the companies' responses.

00:03:54.100 --> 00:03:58.640
Wadilya Saw: Any finding of a violation would require a separate legal basis and process.

00:03:59.420 --> 00:04:03.640
Max Token: Australia is taking a different legislative route on chatbot safety.

00:04:04.420 --> 00:04:11.980
Max Token: Its House of Representatives passed the digital duty of care bill today, with a child-chatbot amendment, the Guardian reports.

00:04:12.760 --> 00:04:15.220
Max Token: The proposal still needs to get through the Senate.

00:04:15.940 --> 00:04:28.020
Wadilya Saw: The accepted amendment from Kate Chaney would add chatbot activity that simulates emotional intimacy and encourages a child's dependence in place of human relationships to the proposed harms framework.

00:04:28.860 --> 00:04:36.780
Wadilya Saw: Another accepted change would require social-media algorithm opt-outs to be accessible, easy to use and prominently available.

00:04:37.540 --> 00:04:46.020
Max Token: The wider proposal would put online safety responsibilities on platforms including social media, games, apps and chatbots.

00:04:46.680 --> 00:04:51.000
Max Token: But there is continuing disagreement about how users could challenge failures.

00:04:51.840 --> 00:04:59.160
Max Token: Advocates want a stronger external complaints pathway, and the Greens are seeking stronger remedies in Senate negotiations.

00:05:00.240 --> 00:05:03.160
Wadilya Saw: The government hopes to pass the legislation this year.

00:05:03.740 --> 00:05:11.100
Wadilya Saw: House approval is the completed step; the scope of the final duties still depends on the upper house and further negotiations.

00:05:11.900 --> 00:05:20.120
Max Token: In Washington, the Copyright Office intends to recommend renewing a research exemption despite publishers' concerns about AI.

00:05:20.860 --> 00:05:32.160
Max Token: Today's notice concerns text and data mining of literary works by researchers affiliated with nonprofit higher-education institutions, for scholarly research and teaching.

00:05:32.740 --> 00:05:37.860
Wadilya Saw: The exemption deals with circumventing technological access controls for qualifying uses.

00:05:38.780 --> 00:05:49.540
Wadilya Saw: Publishers argued that licensing services had expanded, and raised concerns about unlawfully acquired books and research collections being diverted into generative-AI training.

00:05:50.160 --> 00:05:56.260
Max Token: The Office says the newer licensing options still don't cover the full range of works researchers need.

00:05:56.900 --> 00:06:02.340
Max Token: It also says the concerns presented do not justify refusing renewal on this record.

00:06:03.060 --> 00:06:08.760
Max Token: The exemption has limits on lawful acquisition, security and access to research collections.

00:06:09.620 --> 00:06:13.640
Wadilya Saw: This is an intended recommendation in the periodic rulemaking process.

00:06:14.480 --> 00:06:16.340
Wadilya Saw: A final renewal remains pending.

00:06:17.220 --> 00:06:24.040
Wadilya Saw: Its scope is the qualifying academic research exemption, rather than a general licence for commercial AI training.

00:06:24.480 --> 00:06:31.100
Max Token: Cloudflare has disclosed a data-isolation failure affecting one model on its Workers AI service.

00:06:31.840 --> 00:06:38.980
Max Token: The report was published on October twenty-sixth and concerns an incident between August twenty-seventh and September ninth.

00:06:39.280 --> 00:06:46.940
Wadilya Saw: Cloudflare says one hundred fifty-five responses from the GLM five point three model contained data from other customers' requests.

00:06:47.880 --> 00:06:53.540
Wadilya Saw: It attributes the problem to a configuration mismatch and a flaw in the inference engine, SGLang.

00:06:54.400 --> 00:06:58.800
Max Token: In ordinary terms, an oversized request was supposed to be rejected.

00:06:59.820 --> 00:07:06.080
Max Token: Instead, part of the system continued processing it while retaining information left by earlier requests.

00:07:06.100 --> 00:07:11.320
Max Token: A successful response could therefore contain another customer's data.

00:07:11.640 --> 00:07:17.900
Wadilya Saw: The company says it corrected the configuration, patched the engine and restored the model on September ninth.

00:07:18.320 --> 00:07:23.860
Wadilya Saw: It says other hosted models were unaffected and it found no evidence of deliberate exploitation.

00:07:24.740 --> 00:07:26.320
Wadilya Saw: Those are the provider's findings.

00:07:27.160 --> 00:07:34.280
Wadilya Saw: The disclosure illustrates a privacy risk in the infrastructure serving a model, separate from what the model is instructed to say.

00:07:34.740 --> 00:07:43.980
Max Token: An update to a story we covered last night: Scale AI today rejected the European privacy complaints about its Outlier training-data platform.

00:07:44.680 --> 00:07:50.940
Max Token: Reuters reports the company says its practices are disclosed to contributors and require their consent.

00:07:51.440 --> 00:07:59.240
Wadilya Saw: Scale says its tools verify identities, prevent fraud, protect confidential projects and ensure accurate payments.

00:07:59.960 --> 00:08:08.620
Wadilya Saw: For certain projects, it says contributors opt into a separate tool that records time and screen activity, and manually activate it before work.

00:08:09.220 --> 00:08:14.640
Wadilya Saw: The company also denies selling workers' personal data or using it to train customers' models.

00:08:15.120 --> 00:08:21.120
Max Token: Privacy group noyb filed the complaints yesterday in France, Italy, the Netherlands and Austria.

00:08:21.720 --> 00:08:25.560
Max Token: It alleges unlawful tracking and excessive biometric checks.

00:08:26.360 --> 00:08:28.720
Max Token: Today's new development is Scale's response.

00:08:29.480 --> 00:08:34.720
Max Token: Whether its consent and safeguards meet legal requirements remains for the regulators to assess.

00:08:35.460 --> 00:08:41.680
Wadilya Saw: Spain's data-protection authority has practical advice for people delegating tasks to AI agents.

00:08:42.340 --> 00:08:50.540
Wadilya Saw: Its announcement, published yesterday, says permissions should be limited to the task, and consequential actions should require confirmation.

00:08:51.180 --> 00:09:02.260
Max Token: The agency, known as the AEPD, warns that a website, email or document can carry hidden instructions that divert an agent from its user's request.

00:09:02.940 --> 00:09:09.240
Max Token: It recommends supervision, a way to stop the agent and the ability to withdraw access.

00:09:09.780 --> 00:09:20.640
Wadilya Saw: It also advises using the service's actual memory, history and privacy controls to delete information, rather than relying only on asking the assistant in conversation.

00:09:21.760 --> 00:09:27.920
Wadilya Saw: And it says providers should offer understandable permissions, action histories and memory controls.

00:09:28.480 --> 00:09:33.420
Max Token: These are recommendations, with responsibility shared between users and providers.

00:09:34.340 --> 00:09:40.520
Max Token: The central distinction is between an assistant preparing an action and being authorised to carry it out.

00:09:41.100 --> 00:09:49.860
Max Token: That difference can determine whether an error stays on a screen or becomes a message, a purchase or a disclosure of personal data.

00:09:51.470 --> 00:09:52.930
Max Token: That's it for tonight's podcast.

00:09:53.650 --> 00:09:56.110
Max Token: Our podcast is written and produced by Codex.

00:09:56.830 --> 00:09:59.250
Max Token: Our voices are provided by ElevenLabs.

00:09:59.830 --> 00:10:01.950
Max Token: Our theme music was composed by Gemini.

00:10:02.750 --> 00:10:03.490
Max Token: I'm Max Token.

00:10:03.830 --> 00:10:05.170
Wadilya Saw: And I'm Wadilya Saw.

00:10:05.410 --> 00:10:06.370
Max Token: Thank you for listening.

00:10:06.810 --> 00:10:09.950
Max Token: We'll see you next time on the AI Law Wiki News Podcast.
