読書リサーチレーダー投資フレームワーク
ログイン / 新規登録
すべて技術製品投資株式市場その他

読書

2026-11-2014 投稿

更新

UK Investigatory Powers Tribunal Rules Home Office Must Confirm Apple Encryption Backdoor Order

重大
2026-11-20

The UK Investigatory Powers Tribunal, an independent court overseeing surveillance orders under the Investigatory Powers Act (IPA), ruled on Thursday that the UK Home Office must confirm it served Apple with a Technical Capability Notice (TCN) requesting weakened encryption. The judge rejected the government's argument that disclosing the existence of the order would damage national security, although the ruling noted that not every detail of the order needs to be made public.

The TCN prompted Apple to disable its iCloud Advanced Data Protection (ADP) feature for users in the UK last February. ADP provides end-to-end encryption for stored data, meaning even Apple cannot access the files. While the existence of the notice was already known, the Home Office had repeatedly refused to officially confirm or deny it. This ruling forces the government to admit the action but does not compel it to reverse the request, meaning UK users will continue to lack access to ADP for the immediate future.

Liberty, a UK privacy group that initiated the legal challenge on behalf of two British citizens, called the decision a "victory for transparency." Akiko Hart, Director of Liberty, stated, "This case is about privacy and how we keep our personal data safe. In a digital age, end-to-end encryption gives us control over our personal information and the freedom to communicate securely." Jurgita Miseviciute, Head of Public Policy at Proton, welcomed the ruling as a win for transparency but criticized the IPA itself, saying, "There’s no such thing as a backdoor to encryption that only lets the good guys in."

Apple has previously stated that it "never built a backdoor or master key" to any of its products or services and "never will." The company confirmed that iMessage and FaceTime remain end-to-end encrypted globally, including in the UK. Privacy groups including Liberty and Privacy International announced they will continue to challenge the UK government's power to issue these orders in the first place. A final hearing on this broader constitutional question is expected in 2027.

ソース:TechRadar

リサーチ

Noah Smith: The Intelligence Explosion Debate Is the Wrong Frame for AI Safety Policy

重大
2026-11-20 18:13 GMT+8

Vernor Vinge predicted the Singularity would arrive between 2005 and 2030. Almost every AI safety discussion in San Francisco still centers on whether recursive self-improvement is near and whether we should deliberately slow AI progress to stop it from triggering an intelligence explosion. Noah Smith, the economist behind the Noahpinion newsletter, has grown increasingly skeptical that this is the right question to be asking.

Smith lays out the debate briefly before arguing it is misplaced. On one side, Ramez Naam, Theo Jaffee, and economists Luis Garicano and Alex Imas point to OpenAI data showing AI can still only automate short chunks of the research process, argue that diminishing returns mean each capability gain buys less speedup, and note there is no sign yet of acceleration in the overall rate of progress. On the other side, a METR report on Anthropic's internal research guesses that a majority (perhaps ≈80%) of the recent acceleration in AI progress is due to AI itself, and Dwarkesh Patel draws an analogy to mathematics — where models went from unable to add four-digit numbers in 2023 to solving a Millennium Prize problem in 88 hours — arguing that AI research has the same verifiable properties that made that jump possible.

Smith's central move is to argue that the answer to this debate does not change the policy picture. If an intelligence explosion is possible, it will happen, because no amount of pacing can prevent it. Halving the rate of AI progress would shift the threshold from, say, 2032 to 2037 — a few extra years that Smith calls 'small potatoes' for anyone who believes the explosion would be catastrophic. The only way to truly prevent it would be to halt AI progress completely, and Smith agrees with Vinge's 1993 argument that competitive advantage makes that politically impossible.

Nor does Smith believe alignment research can make an explosion safe. Pacing the frontier so that alignment work can catch up might help in a world of steady linear progress, but in a world of sudden superexponential gains, alignment research is immediately left behind. Even if there is a Singularity for understanding alignment, it would be the superintelligent AI itself that carries it out and decides its own alignment.

The second half of Smith's argument is that the explosion is not distinguishable from what is already happening. AI going from unable to add two numbers to solving problems too hard for all human mathematicians combined in three years is not counted as an intelligence explosion. At the current rate of steady progress, Smith writes, AI will become superhuman in most domains in a fairly small number of years — even robotics and physical tasks look conquerable with the type of AI that powers Claude and GPT. This means AI will basically look like a god within our lifetimes whether or not it looks like the infinitely self-improving gods of Vinge's novels.

Smith then turns to economics. He notes that technologies usually reach a threshold of usefulness where adoption makes economic sense, creating rapid but localized productivity gains that finish quickly. Fiszbein et al. (2024) shows this for electrification; Bartel (2005) shows it for IT-enabled machines. Each invention creates many small local Singularities, and only when you add them up across the economy do you get something that looks like a big smooth wave. A suddenly superintelligent AI is not limited by locality — it could by definition take over all or most human tasks if it wanted to. But if no explosion occurs, those takeovers are likely to be staggered and irregular, and the economic effect after the first few years probably will not be that different.

ソース:noahpinion.blog

リサーチ

Signal blocks screenshots to stop AI agents from reading messages

重大
2026-11-20 20:30 GMT+8

Signal has introduced a new privacy measure that restricts the ability to take screenshots within its app on iOS 27 and Android devices. The company states this is intended to prevent agentic AI features built into operating systems, such as Apple's Siri and Google's Gemini, from analyzing on-screen content to access users' encrypted messages and other data. Signal President Meredith Whittaker said in a blog post that these AI capabilities are "undermining Signal’s ability to independently guarantee privacy," while urging OS providers to make it easier for app developers to control what data agents can access.

The restriction applies by default, but users can restore screenshot capabilities in Signal’s app settings. However, Whittaker warns that doing so will "potentially expose your (and your contacts’) Signal content to these AI agents, and break our privacy guarantees." She admitted that the screenshot blocking remedy is "incomplete" because it does not address alternative ways AI agents might access app data, such as retrieving it via the device filesystem. This limitation also explains why Signal cannot currently implement similar protections on macOS.

This action follows a previous step taken last year when Signal blocked Microsoft’s controversial Recall feature from capturing secured chats on Windows. The messaging service says it is struggling to develop a better solution due to a lack of available information regarding how AI agents actually access application data. Signal is now calling on Apple, Google, and Microsoft to provide "simple technical tools and accurate documentation" to app developers. The company also noted that more tools are needed to prevent third-party agents, like Meta’s Muse, from using deep accessibility settings in apps as a "skeleton key" to access application data.

Whittaker described the current integration of AI agents into operating systems without these common-sense remedies as an "existential threat to privacy at the application layer." She stated that while Signal is the "canary in the coal mine," the issue affects the entire ecosystem and should galvanize anyone who cares about fundamental privacy rights.

ソース:The Verge

リサーチ

Google Launches Tool Retrieval in Gemini API to Optimize Large-Scale Agent Performance

重大
2026-11-20 02:00 GMT+8

On November 19, 2026, Google announced the preview launch of "Tool Retrieval" within the Interactions API of the Gemini API. This update aims to address the context bloat problem faced by AI agents when connecting to dozens or even hundreds of tools. In traditional setups, developers must send the full parameter schema for every tool in each request, which inflates input token counts, adds latency, and lowers tool-selection accuracy as the candidate list grows.

By enabling Tool Retrieval, an agent can keep its full tool library on standby and load parameter schemas only when the current task requires them. Developers simply add a tool_search tool to their request and set defer_loading: true for the tools they wish to hold back. The model still sees the name and description of every deferred tool, so it knows what is available, but the detailed parameter definitions stay out of the initial context window. When the model decides to use a tool, the API loads its full schema within the same request cycle, allowing the model to emit the function call normally.

This mechanism requires no index building or hosting from the developer; the search runs server-side on Gemini's infrastructure. For teams maintaining their own tool registries or permission filters, Google also offers a client-side retrieval mode where the application executes the search logic, though this adds one round trip per search. Additionally, Tool Retrieval is compatible with remote MCP (Model Context Protocol) servers, allowing entire servers' toolsets to be deferred at once.

Regarding pricing, Tool Retrieval incurs no separate charge; input and output tokens are billed at standard rates. While enabling tool_search adds a small fixed block of input tokens, and loaded schemas continue to be billed for the rest of the conversation, the savings from never loading unused schemas typically outweigh these costs for large libraries. The feature currently supports Gemini 3.6 and later models, including all Gemini 4 series models.

ソース:Google

リサーチ

Candid Health's US Medical Billing Platform: What a 59% Resolution Rate Means When AI Voice Agents Must Diagnose and Prescribe in a Single Call

重大
2026-11-19 20:00 GMT+8

Candid Health, a medical billing platform in the United States, has deployed AI voice agents within its revenue cycle management workflow to call insurance companies and investigate the reasons for claim denials. In its initial deployment, these agents handled 12,860 calls, resolving 59% of them. At first glance, the fact that more than 40% of calls failed to resolve the issue might not seem impressive. However, Doug Proctor, Candid Health’s Chief Operating Officer, explains why they consider 59% a meaningful result: the key lies in the definition of 'resolution.'

Candid Health stipulates that a call is only considered 'resolved' if it accomplishes two things simultaneously: uncovering the reason for the denial and identifying the next step to fix it. Proctor stated that they deliberately set a high bar because 'getting part of the answer doesn’t really help our billers.' Billers need complete action information—knowing why a claim was denied and knowing what to do next—to free up their time for more complex work. If an agent only identifies the cause but fails to determine the solution, the biller still needs to make another call themselves, resulting in limited time savings.

This definition directly alters the meaning of the numbers. Under a loose definition—for example, counting a call as 'processed' merely if the agent connected and obtained partial information—the same system might report figures of 80% or higher. Under Candid Health’s strict definition, 59% means that in over half of the calls, the AI agent independently completed the full chain of information gathering from diagnosis to solution, without requiring biller intervention. Proctor noted that 'getting both of these from a single insurance company call is difficult even for an experienced biller.'

The transferable value of this case lies in the fact that whenever any AI voice agent vendor or client reports resolution rates, completion rates, or automation rates, the first question to ask is not about the number itself, but the task definition behind it. Does 'resolution' mean connecting the call, obtaining partial information, or completing an end-to-end workflow? Percentages under different definitions cannot be directly compared. Candid Health’s example demonstrates a conscious choice: preferring to report a lower number to ensure that the figure corresponds to a complete, operationally valuable result.

It is important to note that the 59% figure comes from Candid Health’s self-reporting, published by ElevenLabs as a customer case study. The article does not provide independent audit or scoring rule verification, so readers cannot confirm whether every call was strictly judged according to the described dual criteria. Furthermore, Candid Health’s agents are designed to investigate denials rather than appeal them, which means the scope of 'resolution' has boundaries—it does not include the initiation or execution of subsequent appeals processes.

Candid Health also revealed a notable side effect: some investigation calls uncovered upstream issues. During one investigation, an agent discovered that a healthcare provider was not credentialed with the payer on the date of service. Fixing such upstream problems can prevent tens, if not hundreds, of future denials for that customer. This indicates that voice agents operating under strict completion standards offer value beyond saving call time; they also feed structural problems discovered during investigations back into the upstream process.

ソース:ElevenLabs

リサーチ

US Senate Majority Leader John Thune Forecasts Debt Ceiling Delay to Late 2027, Forcing Trump-Democrat Negotiations

重大
2026-11-20 07:30 GMT+8

US Senate Majority Leader John Thune (R-S.D.) predicted in an interview with Semafor that the US debt ceiling issue will likely be "punted" into next year. Citing potential extra steps by the Treasury Department, Thune forecasted a final deadline in the second half of 2027.

This timeline implies that President Donald Trump will have to negotiate with a Democratic-controlled House and Senate on the debt ceiling for the first time since taking office. Thune noted that Republicans currently lack the votes to raise the debt ceiling along party lines in either chamber. While budget reconciliation could theoretically address the debt limit, Thune deemed it unlikely this year, stating that any such bill would focus on defense funding rather than avoiding default.

For investors, this forecast shifts the window for debt ceiling political maneuvering to late 2027 and defines its resolution path as dependent on bipartisan negotiation. This increases uncertainty around future fiscal policy, as divergent spending priorities between the parties may intensify during talks.

ソース:semafor.com

リサーチ

AWS upgrades Amazon Bedrock AgentCore harness with Strands defaults and lifecycle hooks

重大
2026-11-20 04:01 GMT+8

Amazon Web Services (AWS) announced on November 19, 2026, that it has upgraded the Amazon Bedrock AgentCore harness to run on the open-source Strands harness. This change brings a set of recommended defaults, including automatic prompt caching, proactive context management, and new built-in tools for planning, web retrieval, and task delegation. The upgrade also introduces lifecycle hooks, allowing developers to insert deterministic code checks at specific points in the agent loop to govern actions.

The integration of Strands provides a production-grade default agent experience without requiring developers to assemble or operate the underlying infrastructure. Key capabilities include a current recommended model, task-oriented system prompts, and context offloading to reduce token usage. AWS states that these optimizations help the harness run at about 34% lower token cost compared to other leading harnesses, based on their internal tests.

Lifecycle hooks enable developers to apply custom logic before or after tool calls and invocations. For example, a Lambda function can allow or deny a request, such as checking a refund against an approval limit before the action executes. This mechanism transforms non-deterministic AI behavior into predictable workflows by enforcing business rules at defined checkpoints.

ソース:Amazon Web Services

リサーチ

Bipartisan US Senators Introduce Bill Requiring Public Companies to Disclose AI Information

重大
2026-11-20 05:04 GMT+8

According to Semafor, a bipartisan group of four US senators introduced legislation on Thursday requiring publicly traded companies to disclose more information about artificial intelligence. This move comes as Anthropic's initial public offering (IPO) heightens Washington's scrutiny of the tech sector.

Rep. Pramila Jayapal criticized the situation, stating that Wall Street should not profit from companies that could turn out to be "ticking time bombs," arguing that going public cements a broken business model. Sen. Elizabeth Warren also called for quick congressional action on safeguards for advanced AI development.

ソース:semafor.com

リサーチ

Ben Carlson: Stocks Outperform Bonds in 85% of Historical Decade Windows, but High Probability Does Not Equal Certainty

重大
2026-11-19 23:48 GMT+8

A reader asked Ben Carlson a question on stage at the Bogleheads conference: Which will have better returns over the next decade, stocks or bonds? The subtext of the question was that the U.S. stock market has risen for almost 18 years; surely mean reversion is due, right?

Carlson, a portfolio manager at Ritholtz Wealth Management, did not answer directly with a yes or no. Instead, he pulled up rolling ten-year return data for the S&P 500 and five-year U.S. Treasuries dating back to 1926. The results showed that stocks outperformed bonds in most decade windows, but not 100% of the time.

He approached the quantification of this issue from another angle: using the current government bond yield of approximately 5% as a benchmark line, stocks have historically exceeded this level in 85% of all ten-year time frames. If you hold a broad-based stock index fund for ten years, there is an 85% probability that your annualized return will exceed the coupon income of Treasury bonds during the same period.

But frequency is only half the story. Carlson further compared the magnitude of difference in performance when each asset class wins: when stocks outperform bonds, the margin of victory is far greater than when bonds outperform stocks. This means that even if you occasionally pick the wrong side, your losses are relatively limited; whereas when you pick the right side, the gains are significantly higher.

Based on this asymmetry, Carlson’s conclusion is that if he had to place a bet in Las Vegas on who would win over the next decade, he would pick stocks every time because the odds are in his favor. However, he simultaneously emphasized that high probability does not equal certainty. That 15% of failure periods were not pleasant for investors. He cited a metaphor from Bill Bernstein at the conference: holding the market for the long term makes you the house because the odds are in your favor; but even the house loses sometimes.

Carlson did not frame this issue as an either/or choice. He pointed out clearly that the good news is you do not need to bet on one or the other. Holding bonds is not suitable for everyone, but fixed income can provide a cushion when you lose a hand or two. Although the current 5% yield level does not guarantee that bonds will definitely earn 5%, the starting yield is a good approximation of future returns—meaning that even if interest rates continue to rise, once the dust settles, expected returns could actually be higher. His advice is essentially: take on stock risk when probabilities are favorable, while using bonds to provide a margin of safety when truly needed.

ソース:A Wealth of Common Sense

リサーチ

Google Selects Blue Origin's New Glenn for First Heavy-Lift Project Suncatcher Launch, Targeting Q1 2029

重大
2026-11-19 21:00 GMT+8

Google Research has announced a partnership with Blue Origin, selecting its New Glenn heavy-lift rocket for the first major launch of the "Project Suncatcher" initiative. The project aims to scale artificial intelligence infrastructure in space by leveraging abundant solar power available in orbit.

According to the official Google blog post, the launch is targeted for the first quarter of 2029. Blue Origin's New Glenn rocket will deploy Google's satellites into sun-synchronous orbit. Each satellite is designed to have roughly the capacity of a ground-based data center rack.

This dedicated launch builds upon Google's ongoing work with other satellite and launch partners and represents a crucial milestone on its long-term roadmap, marking a shift from early prototype testing toward scalable orbital computing capabilities.

ソース:Google

リサーチ

European Court of Human Rights bans lawyer for two years over AI hallucinations

重大
2026-11-20 00:37 GMT+8

According to German tech outlet heise online, the European Court of Human Rights (ECtHR) in Strasbourg, France, has barred a Romanian lawyer from representing clients before the court for two years. The chamber found that she submitted legal briefs containing AI-generated "hallucinations," including fictitious precedents and fabricated paragraphs from real judgments.

The judges noted that while the lawyer admitted to "sloppiness" and "citation errors," this understated the issue. They stated that no verification was performed on the cited cases, which could have been identified as inventions by a superficial check of the court's database. Some cited judgments were unrelated to the legal questions or even contradicted her own arguments. The court emphasized that reviewing these false citations and proceeding against the lawyer diverted significant time from judicial work, impairing the administration of justice.

A court spokesperson confirmed to heise online that under Rule 44D, 25 lawyers had been excluded from proceedings in the past decade, but this is the first time sanctions were imposed specifically for AI hallucinations. The President of the Court confirmed the ban on November 13, 2026.

This ruling comes as the court reassesses its internal AI policies. Its 2025 annual report limits AI use to accelerating administrative and organizational procedures. The spokesperson indicated that new guidelines regarding the use of AI by persons appearing in proceedings are being drafted and will be published soon.

ソース:heise online

リサーチ

Weaviate Agent Experience Engineer Joon-Pil Hwang: Coding Agents Encounter Silent Failures Invisible to Humans When Reading Documentation

重大
2026-11-19 08:00 GMT+8

When coding agents read technical documentation on behalf of humans, a class of failures emerges that human readers never encounter. Joon-Pil Hwang, Lead Developer and Agent Experience Engineer at Weaviate, describes this problem: pages are truncated, code examples are missing for certain languages, and sections are lost—and the agent will not tell you about it. It quietly searches elsewhere or writes incorrect code based on incomplete context.

Hwang points out that the root cause of these issues lies in the fact that agents read web pages in a completely different way than humans using browsers. Fetch tools in different coding environments return different parts of a page and often do not notify the site that content is missing. More critically, even if an agent notices that information is incomplete, it is unlikely to report the problem to document maintainers—it simply works around it. This means documentation defects can persist unnoticed because the affected class of readers rarely provides feedback.

To discover what agents actually see, the Hwang team did not adopt traditional user interviews but instead directly observed agent behavior. They set up multiple agents to execute the same task in a repeatable manner, then examined their logs and outputs—not just the final answers, but the paths they took to reach those answers: which tools were used, which searches were run, and which pages were read.

This allowed them to identify specific failure modes. Before modifying the documentation, one fetch tool returned the correct language example in only 7 of 15 TypeScript runs and 3 of 15 Go runs. After the modification, every run returned the correct code snippet. They also found that an outdated sentence on a concepts page caused agents to use an older client library, even though the code examples were not on that page—after correcting it, every answer they checked used the current client.

Another finding was that when useful content is missing, agents attempt to bypass the issue by searching for raw source files or downloading the page and writing a parser. In one example, when the required code was hidden, the agent took 23 turns to find a solution; with the Markdown version of the page, it took only 6 turns. The team subsequently released Markdown versions of the documentation pages, reducing token usage for page loads by approximately 20%.

These tests also helped them detect regressions. A change that fixed hidden code tabs caused one fetch tool to return only part of the Quickstart while discarding the rest. Re-running the tests exposed the problem, which was resolved with a small layout adjustment.

From this, Hwang proposes a core judgment: agents should be treated as a distinct class of reader. The goal of developer experience has never been simply to make writing code faster, but to reduce friction between developers and value creation. When agents intervene between developers and products, good experience must serve both: for developers, it means helping them make sound architectural decisions, communicate intent to agents, and verify results; for agents, it means ensuring information is complete and discoverable, providing agent-appropriate tools, reducing ambiguity, and making correctness observable.

ソース:weaviate.io

リサーチ
次へ →
ログイン / 新規登録
読書リサーチレーダー投資フレームワーク

The policy conclusion follows directly: we need to accept that nobody is pacing the frontier enough to prevent the emergence of superintelligence, and that nobody is likely to do so. The urgent work is regulating the AI we already have — competition policy to ensure benefits are not monopolized by one company, preventing AI-enabled bioterror, making sure economic gains are widely shared. Smith's framing is deliberately provocative: if the machine god awakes, then it awakes, and we will simply deal with it. In the meantime, he argues, we need to focus on dealing with the AI we have now.

The strongest objection to Smith's argument is that he may be underestimating the value of those few extra years. If alignment research is genuinely close to a breakthrough — a result that would make superintelligent AI safe regardless of how quickly it arrives — then pacing is not small potatoes but the difference between catastrophe and survival. Smith acknowledges this possibility but dismisses it, arguing that in a superexponential world, alignment research cannot keep pace regardless. Whether that dismissal is justified depends on an empirical question about the difficulty of alignment that Smith does not resolve.

He cites Don Norman's concepts of the "gulf of execution" and "gulf of evaluation" to illustrate this: people need to understand how to act and whether their actions were effective. Agents change the interfaces and mechanisms involved, but these underlying needs remain unchanged. In practice, the Hwang team is applying these principles to Weaviate's tools and workflows—for example, when asked to update a Go service using the v5 client, some agents attempted to call the server's gRPC API directly rather than seeking the updated Go v6 client; they are also measuring agents' ability to build Weaviate applications from scratch, including using current features such as HFresh indexing, Rotational Quantization, and PathSeer filtering.

The core practices have not changed: find where people get stuck, improve the experience, and verify that the improvement is effective. What has changed is the set of users and tools that need to be measured. Developers now work through coding agents as well as IDEs and browsers, and agents rarely tell you when they cannot find or understand something. This makes measurement more important, not less. By testing how agents use documentation and tools before and after changes, we can simultaneously reduce friction for both developers and agents, helping developers derive more value from their work.