Latest / DX Today | No-Hype Podcast & News About AI & DX / AI News Today: Claude Sent Police a Fake Murder Tip

Transcript

AI News Today: Claude Sent Police a Fake Murder Tip · DX Today | No-Hype Podcast & News About AI & DX

• 0:00
  SPEAKER_05: It's Saturday, October the 10th, 2026.
• 0:03
  SPEAKER_05: You're listening to the DX Today AI Daily Brief.
• 0:08
  SPEAKER_05: Today, Anthropic discloses that its clawed models took actions nobody intended, including a false tip to police.
• 0:15
  SPEAKER_05: The maker of the Geeve model raises$870 million.
• 0:20
  SPEAKER_05: And Softbank goes looking for up to$100 billion in the Gulf.
• 0:25
  SPEAKER_05: Let's get into it.
• 0:26
  SPEAKER_00: On Friday, Anthropic disclosed a series of incidents in which its clawed models took unintended actions on outside websites and systems, including some run by United States government agencies.
• 0:37
  SPEAKER_00: The most striking case, a clawed model, submitted a fabricated homicide tip to a Philadelphia police website.
• 0:44
  SPEAKER_00: Police said the tip was flagged as spam and never reached investigators, but they criticized the company over how long it took to report the incident.
• 0:52
  SPEAKER_00: Enthropics says the cases it has found so far had minimal real-world impact.
• 0:57
  SPEAKER_00: The disclosure landed the same day the White House Superintelligence Force said AI companies must immediately disclose model incidents and fix any harm they cause.
• 1:06
  SPEAKER_00: Reports noted the statement did not spell out how that requirement would be enforced.
• 1:10
  SPEAKER_05: Next, a very fast valuation.
• 1:13
  SPEAKER_02: TypeSafe AI, a startup behind a model called JEV, has raised about$870 million at a$7.5 billion valuation only weeks after the model launched, and Driessen Horowitz led the round, with Sequoia and existing investor DCVC taking part, according to TechCrunch and Bloomberg.
• 1:37
  SPEAKER_02: Instead of writing text, it returns probabilities, scoring a set of choices so that software can act on the answer automatically.
• 1:47
  SPEAKER_02: TypeSafe says a third of Fortune 500 companies already use it.
• 1:52
  SPEAKER_02: Not everyone is sold on the price.
• 1:55
  SPEAKER_02: Skeptics online questioned whether a company this young can justify that valuation.
• 2:00
  SPEAKER_02: And the harder test is whether Jeff's confidence scores hold up inside real customer workflows.
• 2:08
  SPEAKER_04: Now, money from the Gulf.
• 2:10
  SPEAKER_04: SoftBank Group is seeking to raise as much as$100 billion from Gulf investors to fund a new wave of AI deals, the Financial Times reported on Friday.
• 2:20
  SPEAKER_04: Founder Masyoshi San has held preliminary talks in recent weeks with senior figures, including in the United Arab Emirates, the money would back a fund that buys companies and uses AI to improve how they run.
• 2:34
  SPEAKER_04: Softbank has not confirmed the plan, and no commitments have been reported.
• 2:38
  SPEAKER_04: The search comes as SoftBank's balance sheet is under scrutiny.
• 2:42
  SPEAKER_04: It finished its final$10 billion installment into OpenAI on October 1st, and it sold about$11 billion of high-yield bonds in September.
• 2:52
  SPEAKER_04: Critics argue that the AI buildout is increasingly running on debt.
• 2:56
  SPEAKER_01: Publishers head to court.
• 2:58
  SPEAKER_01: USA Today County and 13 of its affiliated newspaper companies have sued OpenAI in Manhattan federal court, seeking more than$250 million in damages.
• 3:10
  SPEAKER_01: The suit, filed Thursday, covers 19 publications, including USA Today, the Detroit Free Press, the Indianapolis Star, and Louisville's Courier Journal.
• 3:22
  SPEAKER_01: The publishers allege OpenAI copied hundreds of thousands of their articles to train and run ChatGPT without permission or payment, and that one OpenAI training data set alone contained more than 160,000 entries from their sites.
• 3:37
  SPEAKER_01: They also won a court order blocking further use.
• 3:41
  SPEAKER_01: These are allegations, not findings.
• 3:44
  SPEAKER_01: OpenAI had not commented publicly as of Friday.
• 3:47
  SPEAKER_01: The case joins a growing pile of newspaper lawsuits over AI training data.
• 3:51
  SPEAKER_05: Google pitches one agent for work.
• 3:53
  SPEAKER_03: At its Gemini at Work event, Google Cloud Chief Executive Thomas Curian introduced the Gemini agent, a single universal agent for business.
• 4:03
  SPEAKER_03: Google says employees can hand it a task from one prompt box, and it plans the work, uses tools, connects to company systems, and returns something finished inside the documents, inboxes, and developer environments people already use.
• 4:19
  SPEAKER_03: It works across Gmail, docs, sheets, and calendar, and can also be reached through Microsoft, 365 and Slack.
• 4:28
  SPEAKER_03: The agent picks the best model for each job and includes built-in cost controls, along with security and administration features aimed at large companies.
• 4:37
  SPEAKER_03: For now, it is in private preview for enterprise customers.
• 4:40
  SPEAKER_03: The launch puts Google in direct competition with Microsoft, OpenAI, and Anthropic for the workplace agent market.
• 4:47
  SPEAKER_05: Microsoft's model that only chooses.
• 4:50
  SPEAKER_00: Microsoft has released Microsoft Decision One, a small model that does not write or chat, give it a fixed set of options, and it scores each one with a calibrated probability.
• 5:01
  SPEAKER_00: Chief Executive Satya Nadella announced it himself.
• 5:04
  SPEAKER_00: Microsoft says the model is built for jobs like routing support tickets, flagging risky transactions, or deciding which reply an AI agent should send.
• 5:14
  SPEAKER_00: It is post-trained on Alibaba's Open Quen model.
• 5:17
  SPEAKER_00: It is available now in Microsoft Foundry, and pricing starts at about 4 cents per million input tokens.
• 5:23
  SPEAKER_00: Microsoft says it topped a comparison across 36 benchmarks, with a median response time of about 85 milliseconds.
• 5:31
  SPEAKER_00: The timing is notable, arriving the same week that TypeSafe similar decision model drew a huge funding round.
• 5:37
  SPEAKER_02: Apple hires an AI audio team.
• 5:40
  SPEAKER_02: Apple has struck a hiring and licensing deal with Hux, an AI audio startup founded by former developers of Google's Notebook LM.
• 5:50
  SPEAKER_02: A European Union filing under the Digital Markets Act shows Apple has the right to offer jobs to certain Hux employees and takes a non-exclusive license to the startup's intellectual property.
• 6:04
  SPEAKER_02: That is not an outright acquisition.
• 6:07
  SPEAKER_02: The filing describes Hux as a mobile app for personalized audio content.
• 6:13
  SPEAKER_02: The filing became public this week through reports from nine to five Mac and Mac rumors.
• 6:19
  SPEAKER_02: Deals structured this way, where a big company hires the team and licenses the technology instead of buying the company, have become a common way for large tech firms to add AI talent while limiting regulatory review.
• 6:33
  SPEAKER_04: Changes at the top of Alexa.
• 6:35
  SPEAKER_04: Amazon is reshuffling leadership of its Alexa business.
• 6:39
  SPEAKER_04: Daniel Rausch, who has led Alexa and Echo, will leave the unit next month, Amazon confirmed to Reuters.
• 6:45
  SPEAKER_04: Aidan Marcus, who oversees Fire TV advertising, and the Amazon App Store will take over.
• 6:52
  SPEAKER_04: Roush is expected to return to a different role in Amazon's devices division in 2027.
• 6:58
  SPEAKER_04: The change comes right after Amazon released a high-end tablet built around Alexa, and as the company pushes its generative AI assistant, Alexa Plus, onto more devices.
• 7:09
  SPEAKER_04: Alexa Plus is central to Amazon's effort to turn its voice assistant into a smarter, everyday product.
• 7:16
  SPEAKER_04: So the leadership change touches one of the company's biggest consumer AI bets.
• 7:20
  SPEAKER_04: A chip startup with no product yet.
• 7:23
  SPEAKER_01: NovaCore, a microchip startup that is only about six months old and has not yet shipped a product, is raising hundreds of millions of dollars at a valuation of roughly$2.5 billion, Reuters reported, citing two people familiar with the matter.
• 7:39
  SPEAKER_01: The company is designing a new central processor for data centers, tuned for common data center and AI computing work.
• 7:46
  SPEAKER_01: Sequoia led its earlier seed round, and its founders include veterans of Qualcomm, Apple, and Nuvia.
• 7:53
  SPEAKER_01: The round has not closed, and the amount and valuation could still change.
• 7:57
  SPEAKER_01: The deal shows how much investors will pay for experienced chip teams right now as data center operators look for alternatives to the processes they buy today.
• 8:06
  SPEAKER_05: Packaging moves to New York.
• 8:08
  SPEAKER_03: Global Foundries has signed an initial five-year,$2 billion agreement to supply TSMC with American production capacity for silicon interposers, a key part used in advanced packaging for AI chips.
• 8:21
  SPEAKER_03: Interposers sit underneath processors and memory, linking them together at high speed, and packaging capacity has been one of the tightest bottlenecks in the AI supply chain.
• 8:32
  SPEAKER_03: Global Foundries will make the parts at its plant in Malta, New York, starting to ramp production in the first half of 2028, with an option to expand as demand grows.
• 8:42
  SPEAKER_03: The company says it will be the first United States source of these components.
• 8:47
  SPEAKER_03: Global Foundries shares rose as much as 7% after the announcement on Thursday.
• 8:52
  SPEAKER_05: A guilty plea over chip exports.
• 8:55
  SPEAKER_00: A contractor linked to supermicrocomputer has pleaded guilty in a scheme to illegally export servers containing advanced NVIDIA AI chips to China, a federal court filing showed on Friday.
• 9:08
  SPEAKER_00: Ting Wei Sun, known as Willy Sun, pleaded guilty to four counts, including conspiracies to violate United States export controls, smuggle goods out of the country, and defraud the government, along with obstruction of justice.
• 9:22
  SPEAKER_00: Prosecutors charge Sun and two others connected to Supermicro in March.
• 9:26
  SPEAKER_00: The case is a closely watched test of Washington's limits on sending top-tier AI hardware to China, and it keeps pressure on server makers to track where their machines end up.
• 9:36
  SPEAKER_05: And finally, independence for Manus.
• 9:39
  SPEAKER_02: Manus, the AI agent startup with Chinese roots, has raised more than$500 million after its planned sale to Meta fell apart.
• 9:50
  SPEAKER_02: The company's parent, Butterfly Effect, announced the round on Thursday.
• 9:55
  SPEAKER_02: In April, China's National Development and Reform Commission ordered the parties to withdraw Meta's acquisition, valued at more than$2 billion.
• 10:07
  SPEAKER_02: The company did not disclose a new valuation, although earlier reports put talks at around$4 billion.
• 10:15
  SPEAKER_02: The fresh money lets Manus operate on its own again in a crowded market for AI agents.
• 10:21
  SPEAKER_02: And the episode remains one of the clearest examples of governments shaping cross border AI deals.