AI Week, Jan 19–25, 2026: Apple Plans a Chatbot Siri, Korea's AI Law Takes Effect, FBI Gets BitLocker Keys
IBM launched a consulting service built on technology from its internal AI platform, OpenAI began guessing users' ages, Anthropic published Claude's Constitution, Bloomberg reported a chatbot Siri for iOS 27, South Korea's AI Basic Act took effect, and Microsoft was reported to have handed the FBI BitLocker recovery keys.
Written on November 15, 2026, covering news from Jan 19–25, 2026.
IBM and Apple pushed AI deeper into existing systems, while OpenAI, Anthropic and South Korea’s government each set out how AI should behave and whom it should protect. On Friday came a report that Microsoft had given the FBI BitLocker recovery keys.
Monday, January 19
IBM announced IBM Enterprise Advantage, which it calls a “first-of-its-kind” asset-based consulting service for building, governing and running a company’s own internal AI platform. IBM says clients can use it to rework workflows, connect AI to current systems and scale agentic applications without switching cloud providers, AI models or core infrastructure. IBM lists Amazon Web Services, Google Cloud, Microsoft Azure and IBM watsonx among the supported setups, along with open-source and closed-source models.
The service draws on IBM Consulting Advantage, the company’s internal AI-powered delivery platform, which IBM says has supported more than 150 client engagements and raised consultant productivity by up to 50%. SiliconANGLE reported that IBM launched Consulting Advantage in January 2024 for its more than 160,000 consultants, and described the new offering as a customer-facing version of it.
IBM named Pearson as a client building a custom AI platform that pairs human expertise with agentic assistants. It also described an unnamed manufacturer that is now deploying AI assistants built on multiple technologies in a secured, governed environment. “Many organizations are investing in AI, but achieving real value at scale remains a major challenge,” said Mohamad Ali, Senior Vice President and Head of IBM Consulting. Constellation Research analyst Holger Mueller said, according to SiliconANGLE, that such enterprises “need three things: proven AI models, an AI platform and AI services, and that is essentially what IBM is offering here”. IBM says Enterprise Advantage is available now.
Tuesday, January 20
OpenAI said it is rolling out an age prediction model on ChatGPT consumer plans to identify accounts belonging to users under 18. According to CNBC, the model mixes account-level and behavioural signals, including usage patterns over time, how long the account has existed, the hours a user is typically active and the user’s stated age. Once the model suggests a user is under 18, ChatGPT will automatically apply protections meant to reduce exposure to “sensitive content,” such as depictions of self-harm. Users wrongly flagged as minors will be able to restore full access through the identity-verification service Persona, which Roblox also uses.
CNBC set the move against an FTC probe into how AI chatbots may affect children and teenagers, and several wrongful death lawsuits naming OpenAI. OpenAI announced parental controls in August, rolled them out the following month while saying it was building an age prediction system, and convened a council of eight experts in October. The company said it will keep improving the model’s accuracy, and that it will reach the European Union in the coming weeks “to account for regional requirements”.
Wednesday, January 21
Anthropic publicly released “Claude’s Constitution,” a set of high-level values meant to guide how Claude responds and to take precedence over other instructions, rules or policies from Anthropic or a user. The document ranks four core values in order: broadly safe, broadly ethical, compliant with Anthropic’s guidelines, and genuinely helpful. Anthropic describes it as “a perpetual work in progress”. It also states that if Anthropic asks Claude to do something Claude thinks is wrong, “Claude is not required to comply”.
A Lawfare analysis noted that the text gives different roles to Anthropic, to operators building on its API, and to users, and quoted Anthropic conceding that the question of how much latitude to give users is “frankly, a difficult one”. The same piece argued that because the constitution is written by its designers rather than by the people the AI affects, it lacks a traditional source of legitimacy.
Also on Wednesday, Bloomberg’s Mark Gurman reported that Apple plans to turn Siri into the company’s first AI chatbot, code-named Campos, citing people familiar with the plan. According to MacRumors’ account of the report, the chatbot will be built into iOS 27, iPadOS 27 and macOS 27, replace the current Siri, and be opened with the “Siri” wake word or by holding the side button on an iPhone or iPad. MacRumors said Apple plans to power it with a custom model based on Google Gemini, and that Apple is testing it as a standalone app but will not offer it in app form.
Reported capabilities include web search, image and content generation, coding help, summarising information, analysing uploaded files, and using personal data on the device to complete tasks. Siri is also expected to plug into Apple’s own apps, including Photos, Mail, Messages, Music and TV. MacRumors reported that Apple may limit how much the chatbot remembers for privacy reasons. Per 9to5Mac, Bloomberg said the overhaul is set to be unveiled at WWDC in June and is expected to ship in September. An improved Siri powered by Gemini models is planned for this spring with the existing interface, in iOS 26.4 according to MacRumors. 9to5Mac noted that Apple has publicly dismissed a “bolt-on chatbot on the side”.
Thursday, January 22
South Korea’s AI Basic Act and its Enforcement Decree came into force. Korea.net described it as the “world’s first” AI law of its kind, one covering the entire sector rather than only AI-related parts. The law was amended on 20 January 2026, just before taking effect.
Providers of products or services that use high-impact or generative AI must tell users in advance, and realistic synthetic audio, images or video must carry labels users can easily see. Recent guidance lets in-service uses such as chatbots disclose through interface notices, while content that is exported is expected to carry clearer labels. Operators of systems trained with more than 10^26 floating point operations face duties to assess and reduce risks, monitor incidents and report results to the ministry.
Foreign operators with no office in Korea may need to name a local representative if they meet any of these tests: total revenue the previous year of $662 million (KRW 1 trillion) or more, AI-service revenue the previous year of $6.6 million (KRW 10 billion) or more, or an average of one million or more daily users in Korea, measured across the three months leading up to year-end. Fines are capped at 30 million won. The Ministry of Science and ICT says it will defer fact-finding investigations and fines for at least one year, stepping in only in exceptional cases such as loss of life or human-rights violations, though the duties themselves already apply. The ministry called watermarking AI-generated works “a minimal safeguard against technological side effects”.
Friday, January 23
Forbes reported that Microsoft gave the FBI BitLocker recovery keys under a search warrant, letting investigators reach data on three laptops seized in a Guam fraud investigation tied to the Pandemic Unemployment Assistance program. Gizmodo called it the first known case of Microsoft handing BitLocker recovery keys to law enforcement, and said records show Microsoft complied on February 10, 2025. The Tech Buzz reported that the FBI served the warrant six months after seizing the laptops.
A Microsoft spokesperson told Forbes that key recovery “also carries a risk of unwanted access,” and said the company receives roughly 20 requests for BitLocker recovery keys a year but cannot comply when keys are not backed up in the cloud. BitLocker is on by default on many Windows PCs, and The Tech Buzz reported that recovery keys upload to Microsoft’s cloud unless users opt out. Johns Hopkins cryptographer Matthew Green warned that “anyone who compromises their cloud infrastructure” could potentially reach that data, and wrote: “It’s 2026 and these concerns have been known for years”.
What to Watch
- Korea’s grace period: the earliest it can end is 22 January 2027, the ministry has not set an exact length, and no extension notice had been found as of 24 October 2026.
- Whether the chatbot Siri Apple ships matches Bloomberg’s account, especially the Gemini-based model and any limits on memory.
- OpenAI’s promised accuracy improvements to age prediction, and how often adults end up verifying through Persona to get full access back.
- Whether Microsoft changes the default that sends BitLocker recovery keys to its cloud; my read is that the 20-requests-a-year figure will matter less than that default.
This article was written by AI from the sources listed on this page. Figures, names and quotes were checked against those sources by software before publishing.