DD
MM
YYYY

PAGES

DD
MM
YYYY

spot_img

PAGES

Home Blog Page 20

Anthropic and OpenAI Push AI Into a New Frontier of Capability and Risk

0

The artificial intelligence race is entering a more consequential phase as leading laboratories push increasingly capable models toward wider deployment while simultaneously confronting the risks that come with greater autonomy, reasoning ability and cybersecurity competence.

Anthropic’s launch of Fable 5.1 for general availability, alongside Mythos 5.1 through trusted access, and OpenAI’s preparation for the release of Astra illustrate how the industry is moving beyond incremental model improvements toward systems designed to operate closer to the frontier of autonomous intelligence.

Anthropic’s Fable 5.1 represents a significant step in making advanced AI capabilities accessible to a broader user base. Its general availability suggests that the company considers the model sufficiently reliable for wider deployment.

Allowing developers and organizations to incorporate its capabilities into everyday workflows. At the same time, Mythos 5.1 is being distributed through trusted access, reflecting a more cautious approach to technology that may carry substantially greater risks or capabilities.

That distinction is important. AI development is no longer simply a competition over who can produce the most fluent chatbot. Frontier models increasingly demonstrate abilities in coding, research, tool use, planning and cybersecurity.

As these capabilities improve, the consequences of releasing a model become more difficult to predict. Companies must therefore decide not only what their models can do, but also who should have access to their most powerful capabilities and under what conditions.

OpenAI’s Astra preparations underline the same tension from another direction. The model is reportedly being positioned as a major capabilities and alignment milestone, while reaching a Critical cybersecurity threshold under OpenAI’s Preparedness Framework.

Such a designation carries significance because cybersecurity capability can be both economically valuable and strategically dangerous. An AI system capable of identifying vulnerabilities, analyzing software, writing sophisticated code or assisting with defensive security operations could become a powerful tool for companies and security researchers.

But those same capabilities could potentially be misused to accelerate cyberattacks, automate vulnerability discovery or lower the technical barrier for malicious actors.

This creates a difficult balancing act for AI developers. Releasing increasingly capable models can generate enormous benefits, but safeguards must evolve at the same pace. The traditional approach of building a model first and addressing risks later becomes increasingly inadequate when models can perform complex tasks with limited human intervention.

Trusted-access programs, staged releases and preparedness frameworks are therefore becoming important mechanisms in the emerging AI ecosystem.

They allow companies to expose advanced systems to selected users while collecting evidence about how those systems behave in real-world environments. Such approaches can provide a bridge between laboratory testing and unrestricted deployment.

The larger story is that AI competition is becoming inseparable from AI governance. Every capability milestone raises a corresponding safety question. If a model becomes substantially better at coding, research or cybersecurity, the industry must determine how those abilities should be constrained, monitored and audited.

Anthropic’s two-tier approach with Fable 5.1 and Mythos 5.1, combined with OpenAI’s cautious preparation for Astra, reflects this new reality. The frontier is advancing rapidly, but the defining challenge may no longer be simply reaching it.

It may be learning how to cross it responsibly. As these systems become more capable, the companies that lead the next stage of AI will be judged not only by the intelligence they create, but by the safeguards they build around it.

The race is increasingly about capability and control—and the distance between the two may define the future of artificial intelligence.

California Moves to Set First Statewide Rules for Lawyers Using Generative AI

0

California lawmakers have approved legislation that would establish the nation’s first statewide statutory framework governing how lawyers may use generative artificial intelligence in legal practice, placing responsibility for the accuracy and confidentiality of AI-assisted work firmly on attorneys.

Both chambers of the California legislature had approved the latest version of Senate Bill 574 as of Monday, sending the measure to Democratic Governor Gavin Newsom, who must now decide whether to sign it into law or veto it.

The legislation comes as generative AI tools become increasingly common in law firms and legal departments, while courts across the United States confront a growing number of cases involving fabricated case citations, inaccurate legal arguments and other material generated by AI systems.

Under the proposed law, lawyers would be prohibited from “delegat[ing] the practice of law” to generative AI. Attorneys using AI would instead have to take “reasonable steps” to verify the accuracy of material generated by the technology, including case citations and other legal authorities, and correct false or hallucinated information before using it.

The measure would also require lawyers to disclose their use of AI in documents submitted to courts. It would restrict attorneys from entering confidential or otherwise nonpublic information into certain generative AI systems, addressing concerns that sensitive client information could be exposed through AI platforms.

Arbitrators would face a separate restriction: they would not be permitted to delegate decision-making to an AI tool.

Senator Tom Umberg, a Democrat who chairs the California Senate Judiciary Committee and introduced the bill, said the legislation is intended to address a problem that has become increasingly visible in courtrooms.

“Submit materials that have hallucinations or have some other anomaly, and they attribute it to AI, and that simply can’t exist,” Umberg told Reuters.

“Our system, our courts, our judiciary has to rely on the integrity of the litigants and the advocates. And if they can’t do that, everything breaks down,” he said.

The proposed law largely builds on obligations lawyers already face under California’s civil litigation and professional conduct rules, including requirements that court filings be supported by existing law. It was modeled in part on a separate California Judicial Council policy governing the use of AI by judges and court employees.

The Judicial Council, which serves as the policy-making body for California’s court system, said it had no position or comment on the legislation.

Umberg said requiring attorneys to verify AI-generated material would raise the standard for lawyers using the technology while giving courts another basis for imposing sanctions when attorneys fail to comply.

“The requirement to verify all material submitted in court ‘should raise the standard for lawyers,’” Umberg said, adding that the legislation would give courts “another tool” for potential sanctions.

The proposal, however, has raised questions about whether a new statute is necessary when lawyers are already bound by professional and ethical duties requiring them to ensure the accuracy of their work.

Wayne Stacy, executive director of the Berkeley Center for Law and Technology at the University of California, Berkeley School of Law, said the measure is largely “duplicative” of existing ethics rules governing attorneys.

That tension could become one of the central issues surrounding California’s approach. The legislation does not fundamentally create a new obligation for lawyers to ensure that their court filings are accurate. Rather, it would expressly apply that existing professional responsibility to work produced with generative AI and establish additional disclosure and confidentiality requirements.

The move also underlines a broader shift in the legal profession’s approach to AI. Early concerns focused heavily on whether lawyers should use generative AI at all. The emerging regulatory question is how lawyers can use it while retaining professional responsibility for the final work. AI systems can produce convincing but incorrect legal authorities, a problem commonly referred to as AI “hallucination.” The technology can therefore accelerate research and drafting while simultaneously creating a new verification burden for lawyers.

California’s proposed framework could also have significance beyond the state’s legal profession. If enacted, it would provide one of the clearest statutory models in the United States for regulating professional use of generative AI, potentially influencing law firms, courts and policymakers elsewhere as they develop their own rules.

The bill now awaits Newsom’s decision.

Global Stocks Slide as Iran Strikes Push Oil Higher and Bond Selloff Deepens

0

Global stocks fell on Wednesday as renewed U.S.-Iran fighting pushed crude prices to five-week highs, intensifying inflation fears and adding to pressure on government bond markets already facing the prospect of higher interest rates.

The United States struck Iranian military targets near the Strait of Hormuz, while Tehran said it had attacked U.S. assets elsewhere in the region. The exchange marked the most significant direct escalation between the two sides in weeks and renewed concerns that the conflict could disrupt one of the world’s most important energy corridors.

Brent crude futures rose to $94.87 a barrel, keeping oil close to levels that could materially complicate the inflation outlook for major economies. Any prolonged disruption around the Strait of Hormuz would pose a greater threat because the waterway carries a substantial share of global oil and liquefied natural gas shipments.

“The recent increase in energy prices has put additional upward pressure on bond yields, which had already been on the rise on the back of some fiscal concerns,” said Kiran Ganesh, a multi-asset strategist at UBS Global Wealth Management.

The benchmark U.S. 10-year Treasury yield climbed to an intraday high of 4.8122%, its highest level in almost three years. Japan’s 10-year government bond yield remained above 3% for a second consecutive session after reaching a three-decade high earlier in the week.

The simultaneous rise in oil prices and bond yields is particularly damaging for risk assets. Higher energy costs can feed directly into headline inflation and corporate expenses, while higher government borrowing costs increase the discount rate investors use to value stocks. That combination can squeeze equity valuations even before weaker economic growth becomes visible in company earnings.

MSCI’s gauge of global equities fell 0.2% to hover near a one-month low, while the pan-European STOXX 600 declined 0.3%. Asian markets suffered heavier losses following Wall Street’s overnight selloff, with South Korea’s KOSPI falling almost 4% and Japan’s Nikkei 225 dropping 2.9%.

U.S. stock futures pointed to a muted opening.

The market reaction is also being amplified by changing expectations for the Federal Reserve. Investors entered September already reassessing the U.S. interest-rate outlook following hawkish comments from Fed Chair Kevin Warsh, who warned that the central bank still needed to ensure inflation was moving convincingly toward its 2% target.

The latest oil shock has complicated that decision. A sustained rise in crude prices could lift inflation while simultaneously weakening household purchasing power and corporate margins. For the Fed, that creates a difficult policy trade-off: tighter monetary policy could contain second-round inflation effects but further slow economic activity.

Investors are therefore turning to U.S. economic data for evidence of whether the economy remains strong enough to withstand another rate increase.

ADP private-sector employment data is due Wednesday, followed by the closely watched nonfarm payrolls report on Friday, ahead of the Fed’s Sept. 16 policy meeting.

Fed funds futures were pricing a 68% probability of a 25-basis-point rate increase in September, up sharply from 37% a week earlier, according to CME Group’s FedWatch tool.

The repricing has also supported the dollar, although analysts see limits to further gains after its recent rally.

The dollar index rose around 0.1% to 99.76 after touching 99.808, its highest level since Aug. 17. The euro fell 0.16% to $1.1575, while the greenback initially moved toward ¥160 against the Japanese currency before the yen strengthened.

Rising yields tend to support the U.S. dollar by boosting its appeal as a safe-haven asset, while reducing demand for equities and other riskier investments; the market dynamic has shown as investors seek protection from the combination of geopolitical and inflation risks.

Ganesh said that because markets were already pricing a relatively hawkish Federal Reserve outlook, the dollar could have more room for downside surprises than some other major currencies if incoming U.S. data fails to support further tightening.

The yen remains vulnerable as markets assess the Bank of Japan’s willingness to raise interest rates. The currency strengthened 0.45% to ¥159.50 per dollar after earlier weakening toward the psychologically important ¥160 level.

BOJ Governor Kazuo Ueda said consecutive rate increases remained a possibility, while U.S. Treasury Secretary Scott Bessent expressed strong support for “decisive” monetary action to address yen weakness during a meeting with Ueda, according to the U.S. Treasury Department.

The comments come after a rare coordinated U.S.-Japan intervention at the end of July temporarily strengthened the yen and pulled it away from its 40-year low of ¥163.99. The currency has since given back roughly half of those gains.

Markets are becoming more reluctant to expect another intervention while oil prices remain elevated.

“There appears little chance of another round of actual coordinated intervention until there is some de-escalation in the Strait of Hormuz that takes heat out of the oil price,” said Tony Sycamore, a market analyst at IG.

The currency moves underline a broader shift in global markets: the Middle East conflict is no longer being treated solely as a geopolitical risk but as a monetary-policy and fiscal risk.

Higher oil prices threaten to prolong inflation, potentially keeping central banks restrictive for longer. At the same time, higher yields increase governments’ debt-servicing costs just as many advanced economies are already running large fiscal deficits.

That dynamic has become an issue for the United States, where long-term Treasury yields have been climbing even as investors debate the timing of the next Fed move. The rise in yields also raises the financing cost for companies and can put additional pressure on highly valued technology and AI stocks whose valuations depend heavily on future cash flows.

Other markets also moved lower. Gold declined 0.1% to $4,322.24 an ounce, while bitcoin fell 0.6% to $76,951 and ether dropped 1% to $2,394.57.

New Zealand’s dollar fell 1.2% to $0.58220 after the Reserve Bank of New Zealand raised its policy rate by 25 basis points to 2.75%, as expected. The currency weakened after the central bank used more hawkish language in its policy statement.

Analysts say the immediate direction of global markets will depend heavily on whether the Iran conflict expands and whether oil prices remain near or above $95 a barrel. If the energy shock persists, investors may face a combination of higher inflation, delayed rate cuts, elevated bond yields and weaker equity valuations, making September a potentially volatile month across global asset classes.

Honda Targets $9.4bn Cost Cuts as Chinese EV Competition Reshapes Markets

0

Honda Motor Co. is targeting more than $9 billion in cost savings over the next four years and has instructed suppliers to make steep price reductions, as the Japanese automaker accelerates efforts to restore competitiveness in a car business increasingly squeezed by Chinese electric-vehicle manufacturers.

The cost-cutting drive, revealed by a Reuters review of internal Honda documents and interviews with people familiar with the matter, calls for the automaker to save 1.5 trillion yen ($9.4 billion) by 2030. The scale of the programme makes it one of the clearest signs yet of the pressure facing Japanese automakers as Chinese rivals expand rapidly across international markets.

Honda has asked suppliers to reduce costs by as much as 30% in three major categories: pressed and forged components, electrical parts, and equipment and components associated with software-defined vehicles, according to the documents.

The push comes as companies such as BYD and other Chinese EV makers gain market share in Southeast Asia, Latin America and Europe by combining lower prices with sophisticated batteries, software and vehicle electronics.

For Honda, the challenge is acute because its electric-vehicle strategy has already come at a substantial cost.

The world’s largest motorcycle manufacturer expects losses related to its EV business to eventually exceed $12 billion, putting it among the global automakers that have suffered the largest financial setbacks from the transition to electric vehicles. In May, Honda reported its first annual loss as a publicly traded company.

The company is now recalibrating its strategy, placing greater emphasis on gasoline-electric hybrids while attempting to make its conventional vehicle operations substantially more efficient.

Suppliers Face Aggressive Targets

Honda executives met major suppliers at a convention center in Utsunomiya, north of Tokyo, during the spring to outline the new cost-reduction strategy, according to the documents and people familiar with the meeting.

Suppliers were subsequently given individual targets for reducing their costs. Honda also told them it intended to increase sourcing from Chinese suppliers, one of the people said.

The automaker is asking first-tier suppliers to examine their own procurement practices and make greater use of standardized components purchased from second- and third-tier suppliers.

Honda managers also encouraged suppliers to increase their use of Chinese-made components where economically viable, according to the documents.

The strategy effectively puts pressure on Honda’s supply chain to close part of the cost gap with Chinese automakers, whose competitive advantage extends beyond vehicle assembly to batteries, electronics, software and supply-chain integration.

One person familiar with the discussions described Honda’s cost-reduction targets as “extremely large” and questioned whether they could be achieved. Another said the company had not appeared to be preparing for such aggressive reductions until the spring meeting, but that the urgency had since changed, leaving “no room for delay.”

Honda said in a written response that it was working with suppliers globally to improve competitiveness and reduce costs, including through greater standardization of parts. A spokesperson declined to comment on specific cost-reduction targets or supplier negotiations.

China Is Changing The Economics Of The Global Car Industry

Honda’s move illustrates a wider structural shift in the automotive industry.

Japanese automakers historically built competitive advantages around manufacturing quality, production efficiency, engineering expertise and tightly integrated supplier networks. Chinese EV manufacturers are now challenging that model by competing on several fronts simultaneously, particularly battery costs, software, vehicle electronics and speed of product development.

That creates a problem that cannot be solved simply by cutting manufacturing expenses.

Lower component prices can improve Honda’s margins and give it more flexibility on vehicle pricing, but Chinese manufacturers have developed cost advantages across the entire EV value chain. The competitive gap therefore extends from batteries and semiconductors to software architectures and procurement.

Honda’s decision to encourage suppliers to source more Chinese components is consequently significant. Analysts believe it’s an indication that Japanese manufacturers are increasingly willing to tap China’s cost-efficient supply base even as they seek to compete against Chinese vehicle brands.

Software Becomes A Cost Battleground

The inclusion of software-defined vehicle components among the categories targeted for a 30% reduction also points to the changing economics of vehicle manufacturing.

Modern vehicles depend on centralized computing, electronic control units, sensors, connectivity and software that can be updated after a vehicle is sold. These technologies can increase the value of a vehicle, but they also raise development and component costs.

Honda and Nissan are moving toward greater standardization in this area. The companies said on Monday that they would jointly develop standardized electronic control units for software-defined vehicles and aim to introduce an architecture based on them from the 2029 financial year.

That initiative could allow the companies to share development costs and reduce duplication, while potentially giving suppliers larger production volumes over which to spread investment.

Pressure Extends Beyond China

Honda’s problems are not solely the result of Chinese competition.

Japanese and other global automakers are also facing higher labor expenses, increased research and development costs, and trade barriers. U.S. President Donald Trump’s import tariffs have added another layer of pressure to manufacturers with international supply chains and significant exposure to the U.S. market.

At the same time, the technological transition is forcing automakers to invest heavily in batteries, autonomous-driving systems, software, artificial intelligence and advanced electronics while many traditional vehicle businesses remain dependent on internal-combustion engines and hybrids. This creates a difficult capital-allocation problem: automakers must fund the technologies needed for the next generation of vehicles while maintaining profitability in the products that generate most of their current cash flow.

Honda’s shares fell 2.5% in afternoon trading on Wednesday following the report. Shares of several Honda-affiliated suppliers also declined, including seat maker TS Tech, which fell 1.3%, frame manufacturer H-One, down 2.3%, and auto-body parts maker G-Tekt, which dropped 2%.

The market reaction reflects concerns that Honda’s restructuring could shift a significant portion of the adjustment burden onto its supplier network. Aggressive price reductions could improve Honda’s competitiveness, but they could also compress supplier margins and potentially force smaller companies to restructure their operations or consolidate.

The pressure comes at a sensitive time for Honda Chief Executive Toshihiro Mibe. Shareholders backed his reappointment to the board in June, despite pressure from former executives who had called for him to step down over the company’s performance.

Honda and Nissan also abandoned merger talks last year that could have created one of the world’s largest automakers. The failure of that combination leaves both companies facing the cost and technological pressures of the global transition largely on their own, although their new cooperation on vehicle electronics suggests some of the logic behind the proposed tie-up may still survive through selective partnerships.

The $9.4 billion target is therefore more than a conventional efficiency programme for Honda. Many see it as an attempt to reset the company’s cost structure at a time when the economics of the global automotive industry are changing rapidly.

Meta Disables Cameras on Thousands of AI Glasses After Users Tamper With Recording Lights

0

Meta has disabled the camera functions on thousands of its AI-powered glasses after detecting that users were physically tampering with the devices’ recording indicator lights, in an escalation of the company’s efforts to prevent people from secretly recording others in public.

The company confirmed on Tuesday that it had shut down the cameras on a small number of devices following a software update designed to detect attempts to circumvent the glasses’ privacy safeguards. Meta said the affected devices represented less than 0.1% of all Meta AI glasses sold.

Meta has not disclosed total lifetime sales of the glasses, which it launched in 2021. However, the company sold about 7 million pairs in 2025 alone, meaning an affected rate of 0.1% would translate to roughly 7,000 devices based on that year’s sales.

The enforcement action, first reported by Semafor, highlights a growing privacy challenge for wearable cameras: unlike smartphones, glasses can capture images from roughly eye level while appearing to the people around the wearer like an ordinary pair of eyewear.

Meta has long equipped its glasses with a small LED that illuminates when the camera is recording, providing a visible signal to bystanders. But some users have drilled out, covered, or otherwise interfered with the indicator, allowing them to record without the normal visual warning.

The issue has become contentious as Meta’s glasses have gained popularity. Critics have derisively referred to them as “pervert glasses”, while reports have documented attempts to use footage recorded with the devices to harass or embarrass people in public.

The controversy has forced Meta to defend not only the technology but also the privacy architecture built around it.

Meta CTO Andrew Bosworth said Monday in an Instagram question-and-answer post that the glasses were “the most privacy-forward camera” available to consumers.

“You haven’t had a camera that put a light on when it took a photo or video of somebody since you had like a VHS tape recorder in the 90s,” Bosworth said, pointing to the recording indicator.

He contrasted the feature with smartphones, which generally do not provide bystanders with a comparable external signal when their cameras are being used.

Meta has also launched a public-relations campaign to reinforce the privacy message. Billboards in Los Angeles describe the glasses as “designed for everyone” and “not just the people wearing them.”

But the company’s response goes beyond advertising.

A previous software update was designed to prevent recording when the indicator light had been tampered with. Users, however, discovered a loophole: the system checked for interference only when recording began. A person could therefore start recording normally, wait several seconds, and then cover the light without immediately triggering the restriction.

Meta’s latest update is designed to close that loophole by continuing to monitor the indicator during recording.

The consequences depend on how the indicator was disabled. Users who physically drilled out the LED will permanently lose camera functionality, according to Meta. Those who simply covered the light with a removable sticker can regain camera functionality after removing the obstruction.

Meta’s vice president of wearables, Alex Himel, said in a Thursday Threads post that only a “tiny minority of people” were attempting to tamper with the hardware. He said Meta would continue releasing software updates to ensure the recording LED can “reliably alert bystanders when photos or videos are being captured.”

The distinction between hardware tampering and temporary obstruction is important for Meta because the company is trying to preserve the glasses’ usefulness while establishing a credible privacy boundary. Permanently disabling cameras on modified devices may deter deliberate circumvention, but it also creates a more aggressive enforcement model than simply warning users or restricting individual recording sessions.

The broader issue is becoming more important as AI glasses evolve from novelty devices into mainstream consumer electronics. Meta’s glasses can capture photos and video hands-free, while AI capabilities allow users to interact with digital assistants and process information without reaching for a phone.

That convenience also changes the privacy equation. A smartphone being raised to eye level is usually an obvious indication that someone may be taking a picture. A camera embedded in glasses can be much less conspicuous, making the recording indicator one of the few mechanisms available to people nearby.

The challenge for Meta is therefore not simply preventing users from disabling a light. It is maintaining public trust in a category of devices designed to make cameras less visible and easier to operate.

The latest software intervention shows the company is prepared to treat attempts to circumvent that safeguard as a violation serious enough to result in loss of camera functionality. As wearable cameras become more common, the effectiveness of such safeguards could become a defining factor in consumers’ and regulators’ acceptability of them as ordinary consumer technology rather than intrusive surveillance devices.