Reading

US-based AI adoption measurement startup Larridin has announced the completion of a $30 million Series A funding round. The round was led by Pelion Ventures, with participation from existing investors Andreessen Horowitz (a16z) and Gradient Ventures, as well as new investors GFT, W23, Thayer, DG Daiwa Ventures, Cougar Capital, and Alumni Ventures. This financing brings the company’s total capital raised to $47 million.

Larridin focuses on providing an "AI value measurement platform" designed to solve the problem of enterprises increasing their spending on artificial intelligence tools while struggling to prove their actual business return. As companies move beyond simple chatbots to using autonomous AI agents for workflows, traditional metrics like seat licenses or login counts no longer reflect true impact. Larridin links data on AI tool adoption with employee productivity metrics and expenditure data, helping leaders identify where AI is driving value and where spend is being wasted.

According to SiliconANGLE, Larridin co-founder and CEO Russ Fradin pointed out that many organizations suffer from evaluation blind spots. For instance, one customer survey showed 92% of engineers were active AI users, but deeper analysis revealed that a single engineer accounted for 65% of the company’s entire AI budget. Additionally, Larridin’s newly released "Engineering AI Impact Report" found that while the proportion of AI-assisted pull requests increased, median code review wait times also rose due to downstream bottlenecks, failing to deliver expected efficiency gains.

US-based artificial intelligence company OpenAI announced on November 2, 2026, a significant increase in rewards under its Security Bug Bounty program. According to the official announcement, the midpoint of reward ranges for medium, high, and critical findings has been more than quadrupled on average. A critical vulnerability in a core product such as ChatGPT, Codex, or the API Platform will now earn up to $30,000. Exceptional critical findings on those targets remain eligible for up to $100,000, and now up to $172,000 with bonuses.

In addition to base reward increases, OpenAI introduced new incentive structures to encourage high-quality reports. These include a "high-quality report bonus multiplier" for submissions that provide a reliable proof of concept, isolate the bug, and demonstrate its impact; and a five-tier "loyalty program bonus" designed to support researchers who consistently partner with the company. Furthermore, "priority campaign multipliers" will offer short-term high-reward opportunities for early testing of newly launched features.

OpenAI stated that these updates are designed to further encourage independent researchers to pursue vulnerabilities that internal testing may have missed, continuing to build a collaborative, community-first security environment. The company also noted that retroactive payments will be made for bounties awarded in the 30 days before this announcement if the finding would have earned more under the new ranges. This move reflects the growing importance of frontier AI models in cybersecurity and a strategic shift towards using economic means to accelerate external defensive participation.

Voice cloning detection faces a fundamental difficulty: no single acoustic metric can independently determine whether a recording is genuine or fake. Shafik Quoraishee, who has a background in acoustic analysis, conducted a controlled experiment on himself, the results of which illustrate exactly why this problem is so thorny. He recorded twenty-four sentences, cloned the same content using a free open-source tool, and then compared the jitter (tiny variations in the length of vocal fold vibration cycles) and shimmer (tiny variations in the height of vibration peaks) of the two sets of audio.

Sentences recorded in a room with a heater running showed that the cloned version's jitter was statistically indistinguishable from his real recordings, but its shimmer was steadier than the real voice. In a quiet environment, however, the cloned voice actually betrayed itself due to having too little background noise—all twelve sentences displayed signal characteristics cleaner than the real voice, even after undergoing degradation simulating a telephone call. In other words, background noise happened to mask the differences in jitter between the clone and the real voice, while the quiet environment exposed the clone's flaw of lacking background noise. Each condition leaked different information, but no single metric could cover all situations simultaneously.

Sarah Barrington, a doctoral candidate at the University of California, Berkeley, who researches AI voice cloning, further points out that this difficulty runs deeper than the experimental data suggests. AI detection models can be tuned to detect the signatures of one cloning company's tools, but this ability often does not transfer well to tools from other providers. Spectral analysis can reveal traces specific to certain cloning tools, but there is no single universal measure that works for all of them. "You would expect to immediately see giveaways," Barrington says. "That is absolutely not the case."

One of the most promising approaches currently may be embedded within the cloning technology itself. Companies such as Google hide inaudible watermarks in generated audio—these changes are extremely subtle, imperceptible to the human ear, but readable by detectors, and they travel with the audio file wherever it is shared. However, watermarks are not foolproof: sufficiently heavy post-processing can erase the markers, and watermarking features are optional in some open-source tools. Barrington compares a good watermark to a locked door, saying it can keep out 99 percent of bad cases, while the remaining 1 percent must rely on audio forensics techniques evolving alongside the cloning tools.

Telephone calls further weaken detection capabilities because they compress or degrade key forensic details in the voice, making clones even harder to spot. Quoraishee's experiment in a quiet room still revealed differences even after simulated phone processing, but this was only the result of his personal voice and a single open-source tool; commercial systems used in actual fraud may produce entirely different acoustic characteristics. In the ongoing race between detection and cloning, neither side currently holds a definitive advantage.

Researchers at the Massachusetts Institute of Technology's Computer Science and Artificial Intelligence Laboratory (MIT CSAIL) published a blog post on November 2 introducing a method called "Weight-Space Programming." This approach allows the frontier large language model GPT-6 Astra to directly read and modify the weight parameters of a smaller model, Qwen3.5-0.8B, without using gradient descent or traditional training processes.

According to the blog, Astra analyzed Qwen and found it was prone to repetitive reasoning loops in mathematical tasks. By locating specific output columns in layers 15 and 19 and modifying only 0.016% of the weights, Astra removed this bottleneck. In the MATH-500 benchmark, Qwen's score rose from 7.9% to 48.1%, an increase of approximately six times; instruction-following capabilities also improved threefold.

The study also demonstrated Astra's ability to generate network weights from scratch. In the MinAtar Breakout game, a network generated by Astra in a single response scored 517.4, significantly outperforming a baseline reinforcement learning model trained for 500 million steps (which scored 25.3). However, the authors note that this method is currently primarily used to "remove obstacles hindering capability" rather than granting the model abilities it did not originally possess. For instance, applying the same method to Gemma 3 270M, which rarely exhibits looping issues, did not result in performance gains.

It must be emphasized that all reported results are self-tested data from the research team and have not yet been independently reproduced.

Exclusive reporting by tech media outlet Semafor indicates that Fox Corporation's $22 billion acquisition of streaming platform Roku is expected to receive approval from the U.S. Department of Justice (DOJ) in its antitrust review, likely through a "consent decree." This development comes despite efforts by Chris Ruddy, owner of right-wing media network Newsmax and an ally of President Donald Trump, to block the deal.

It remains unclear what specific terms the consent decree would entail or what conditions are being considered. People familiar with the matter noted that the DOJ could still change course. Both Fox and Associate Attorney General Stanley Woodward declined to comment on the deal. In a statement, Woodward said: "However, we can affirm that this DOJ under President Trump’s leadership will continue to prioritize affordability for all Americans across our economy."

The transaction would unite Fox’s empire of news and sports content with Roku’s dominant connected-TV operating system. Rivals have expressed concern that Fox might favor its own content. Roku’s free, ad-supported channel also competes for viewers and content with Tubi, which is owned by Fox. Fox CEO Lachlan Murdoch has stated his "expectation" is to run the two businesses separately while selling advertising across them.

Semafor first reported in September that antitrust regulators were widening their inquiry into the deal, days before the two companies confirmed the deeper probe, known as a "second request." That enhanced scrutiny came in part because of pressure from Newsmax majority owner Chris Ruddy, who bills himself as a Trump confidant and has had a long-running spat with Fox owner Rupert Murdoch. Murdoch’s camp has previously dismissed Newsmax’s efforts as a bid to "sue their way out of their own competitive failures," but it is clear that Ruddy has been a headache for the Murdoch clan. Investors have fretted about the possibility that Fox’s deal for Roku could get tied up as a result of Ruddy’s efforts.

According to TechRadar, Palantir cofounder Joe Lonsdale is reportedly supporting Safer Human Medicine, a startup planning to establish a large-scale macaque breeding facility in the United States that would eventually house 30,000 monkeys. The initiative aims to address the supply pressure on animal test subjects caused by AI-accelerated drug discovery.

AI technology enables researchers to generate potential drug candidates more rapidly, increasing the need for toxicity, safety, and biological effect testing. Macaques are frequently selected due to their biological similarities to humans. However, US laboratories face severe shortages after China banned exports of lab monkeys in 2020 (previously accounting for about 60% of US imports) and alternative sources like Cambodia faced allegations of wild-caught animals being misrepresented as captive-bred. Reported prices for macaques have surged from approximately $3,000 in 2019 to over $20,000 today, a 700% increase.

Safer Human Medicine plans to gradually expand its population over approximately 20 years, starting with an initial stock of 500 to 1,000 macaques obtained from Asian facilities. However, female macaques typically reach reproductive maturity around four years of age and produce one offspring annually, meaning domestic capacity will take a long time to scale. Additionally, the plan has attracted criticism from animal welfare advocates, potentially leading to legal challenges that could further delay the project.

On October 30, space intelligence company InSpatio launched China's first embodied simulation data collection center in Hangzhou. According to InSpatio, the facility is projected to produce 1 million hours of simulation training data per month, reducing the comprehensive cost per data unit to approximately one-tenth of traditional real-machine teleoperation methods. This initiative aims to address the persistent shortage of high-quality training data in the field of embodied intelligence.

Currently, data acquisition for embodied AI relies on two main paths: physical world collection via real-machine teleoperation and first-person view (EGO) wearable devices, and digital world generation through 3D modeling and physics simulation. However, both face efficiency bottlenecks. Physical collection is limited by time, site access, and environmental adjustment flexibility, often yielding less than half an hour of effective trajectory per operator-hour. Traditional simulation modeling depends on manual scene construction, taking days or weeks, and struggles to maintain geometric and physical consistency, leading to the 'Real2Sim Gap' during transfer to reality.

InSpatio's proposed solution is to build an 'Embodied Token Factory.' Using its proprietary Topos spatial intelligence base, the system converts real-world spaces into interactive digital scenes through three automated industrial lines. Upstream, portable scanning or native 3D generation captures over 100 square meters of space in 15 minutes. The system then performs object decomposition, blind-spot completion, and physical calibration to transform these scans into runnable digital masters within hours.

Midstream, once scenes are digitized, the system can launch multiple parallel simulation channels in the cloud. The underlying physics engine calculates mechanical results in real-time, filtering out failures to batch-generate embodied tokens containing depth, pose, contact force, and action semantics. This shifts data production volume from reliance on manual on-site setup to dependence on digital scene reusability and cloud computing scale.

Downstream, InSpatio implements deduplication standards and tiered data delivery based on spatial geometry, physical interaction, and task closure evaluation. Data production also forms a closed loop with model testing, allowing targeted generation of complex environments to address specific weaknesses identified in client models. While this industrial process demonstrates scalability potential, the actual effectiveness of the generated data and its contribution to narrowing the Real2Sim Gap remain subject to future independent verification.

Many open source maintainers complain about the influx of AI-assisted pull requests (PRs): large language models are generating code submissions that flood repositories, significantly increasing the review burden. Simon Willison, author of the open-source tool llm, discusses his practical experience with this issue in the release notes for version 0.37—he states that llm is the project most affected by this trend, but he does not consider it a problem.

His response operates on two levels. On the tooling level, he uses Opus 5.5 or GPT-6 Astra to assist with triage, sifting through numerous submissions to identify the most valuable fixes. On the cognitive level, he no longer views AI-assisted PRs as code submissions requiring line-by-line review; instead, he treats them as bug reports with attached examples—sometimes landing them directly, sometimes implementing his own fix.

This release includes fixes from 12 contributors, almost all of which were AI-assisted. Willison notes that the review burden was far less than he had anticipated.

Dutch cybersecurity startup Hadrian announced it has completed a $40 million Series B funding round, bringing its total raised to $65 million. The company plans to use the new capital to expand operations in Europe, the Middle East, Africa, and the United States, as well as to increase investment in engineering and research.

Founded in Amsterdam in 2021, Hadrian uses AI agents to simulate attacker techniques, helping enterprises inspect systems and applications exposed to the public internet. Its External Attack Surface Management product, Atlas, continuously inventories assets and performs security tests when new assets are discovered or existing ones change. When new vulnerabilities are published, Atlas compares software versions to identify affected assets and tests whether the vulnerabilities can be exploited.

In March this year, Hadrian launched Nova, an AI-agent penetration testing product. Nova allows enterprises to specify the applications and scope for testing, with multiple AI agents attempting to chain different vulnerabilities to assess potential attack paths and their impact. According to Hadrian, most tests can be completed within 24 to 48 hours. UK cybersecurity media outlet Cyber Insider noted that while automated penetration testing helps identify issues, enterprises must still accurately manage external assets and control the testing process to ensure security risks are genuinely reduced.

Apple Inc. released its financial results for the fiscal fourth quarter ended September 26, 2026, on November 2. The company reported total revenue of $114.477 billion, a 12% increase year-over-year. Net income stood at $30.711 billion, with diluted earnings per share (EPS) at $2.10, up 14%. Gross margin reached 47.9%, benefiting approximately one percentage point from tariff refunds.

Performance varied across segments. iPhone revenue hit a record for the period at $57.583 billion, while Mac revenue was $10.792 billion. Notably, Services revenue reached an all-time high of $31.573 billion. iPad revenue was $6.229 billion, and Wearables, Home and Accessories generated $8.3 billion. Greater China revenue rose to $17.539 billion from $14.493 billion in the same period last year.

This marks the first quarterly report under new CEO John Ternus. During the earnings call, Ternus stated that Mac sales hit an all-time high this year and highlighted efforts to commercialize AI capabilities. Some server-side AI features will have usage limits to encourage higher-demand users to subscribe to iCloud+. Ternus repeatedly referenced the new Siri, calling it "the Siri we've always wanted."

Management provided optimistic guidance for the next quarter, projecting total revenue to grow 7% to 10% year-over-year in the December 2026 quarter, setting a new record for quarterly revenue. Services revenue is expected to grow 8% to 9%.

The U.S. Census Bureau has made a significant wording change to a key question in its bi-weekly Business Trends and Outlook Survey (BTOS). This shift stems from Executive Order 14434, signed by the U.S. President, which aims to inaugurate an era of "Super Intelligence."

According to an analysis of the latest survey documents by developer Simon Willison, starting with the collection cycle on October 19, 2026, the question regarding artificial intelligence usage was changed from "In the last two weeks, did this business use Artificial Intelligence (AI)..." to "In the last two weeks, did this business use Super Intelligence or SI (formerly referred to as Artificial Intelligence or AI)...". The examples provided also shifted from general terms like machine learning and natural language processing to politically charged phrases such as "SI-powered data analysis" and "generative SI for marketing."

This change directly impacts data comparability. Previously, the question had been broadened in November 2025 from "producing goods or services" to "any business functions," causing the percentage of affirmative responses from large companies to jump from around 12% in September 2025 to 30% by February 2026, and reaching approximately 40% by October 2026. However, with the introduction of the new "Super Intelligence" wording, currently published data still reflects the old "Artificial Intelligence" definition, while the first results incorporating the new wording are expected to be released the following Thursday. This means that in the short term, the official tracking of U.S. corporate AI adoption via BTOS will suffer from definitional discontinuity or gaps.

For investors and researchers monitoring AI industry penetration and macroeconomic impact, this event highlights how non-market factors can distort core statistical indicators. When official definitions are replaced by politicized terminology, historical trend lines may become invalid, and new baselines require time to establish. Readers should treat recent conclusions drawn from BTOS data about "accelerating" or "slowing" AI adoption with caution, awaiting the release of the first batch of data under the new criteria to recalibrate expectations.

Match Group announced during its third-quarter 2026 earnings call that it is testing a cheaper subscription plan for its dating app Hinge. The move aims to convert more users into paying customers as people grow increasingly tired of expensive dating app subscriptions. Steve Bailey, Chief Financial Officer of Match Group, stated the new tier is designed for women and price-sensitive users, offering features like additional filters and custom location boundaries.

The decision highlights a strategic tension within Match Group’s portfolio. According to Bailey, Hinge is one of the company’s highest-priced products, with revenue per payer at approximately $30—nearly twice that of Tinder. However, Hinge’s payer penetration is only slightly more than half that of Tinder. While Hinge reported strong growth in Q3 2026, with global monthly active users up 12% year-over-year and direct revenue increasing 19%, its paid subscriber count stands at 2.3 million. In contrast, Tinder generated $480 million in direct revenue but saw a slight 2% decline year-over-year, with 8.9 million paid users.

This shift marks a reversal from Hinge’s previous strategy. In 2023, the company experimented with a more expensive tier aimed at users willing to pay extra for advanced features. Now, facing broader subscription fatigue, Match Group is exploring whether a more affordable option can appeal to a wider audience without diluting the brand’s premium positioning. The exact price of the new tier remains undisclosed.