The approved increase affects Central and Wan Chai crossings to Tsim Sha Tsui. Weekday fares rise by HK$0.60 to HK$1; weekend fares by HK$1 to HK$1.80.

The next Star Ferry fare increase starts on 25 January 2027. There is no new charge to budget for on this weekend’s crossing. The approval covers the Central-Tsim Sha Tsui and Wan Chai-Tsim Sha Tsui routes.

Single trips will cost HK$0.60 to HK$1 more on weekdays and HK$1 to HK$1.80 more at weekends. Adult fares will be capped at HK$6 on weekdays and HK$8.30 at weekends. The monthly pass stays at HK$190.

That leaves a useful distinction for passengers: regular monthly-pass users keep the same bill, while people paying trip by trip will face the increase. The government said it kept weekday increases below weekend increases and held the pass price to ease the burden on regular commuters.

The Executive Council approved the application on 13 October; the announcement followed on 21 October. Star Ferry had sought increases of HK$0.80 to HK$2, which the government moderated.

There is still time to plan the next cheap harbour outing. Put the January date in the budget, then check the fare for the route and day you actually travel.

Open the page → · show less

Sansec warns that repairing the hosted analytics script can leave shoppers exposed, while earlier card payments require an immediate response.

A shopper returning to a Convertr checkout after its repair could still run the malicious analytics script saved in their browser, according to security firm Sansec. The server-side cleanup has a browser-side remainder. Another payment can still encounter malicious code.

In its October 27 report, BleepingComputer carried Sansec’s confirmation that Convertr had “removed the skimmer and upgraded Plausible to a patched version.” The same report warned that browser caching could keep the malicious script active for up to 24 hours after cleanup.

The script arrived as analytics. Attackers modified a shared Plausible tracker hosted on Convertr’s infrastructure, which loads on checkout pages across merchants using the Brazilian e-commerce platform. That gave the injected code a place beside the shopper’s payment fields.

Sansec identified the campaign on October 24. By October 26, the researchers say, the code specifically targeted Convertr’s payment field names. It was configured to steal card numbers, expiry dates, card security codes, cardholder names and Brazilian taxpayer identifiers. That describes its collection instructions. The report does not establish which fields were actually taken from each shopper.

The suspected entry route also needs its qualifier. Sansec suspects CVE-2026-8467, an unauthenticated command-execution vulnerability in Convertr’s outdated Plausible Community Edition installation. That route remains unconfirmed.

Recovery has separate consequences elsewhere in this week’s reporting. GMO confirmed publication of company and customer data after its Onamae Desktop Cloud intrusion, while BleepingComputer reported that the threatened ransomware activation apparently did not occur. A machine escaping encryption can still leave its owner dealing with stolen information.

For Convertr shoppers, Sansec’s immediate advice is specific: anyone who paid by card since October 24 at a Brazilian store whose checkout address contained “checkout.convertr” should contact their bank immediately and ask to block the card. Clearing a browser cache cannot recover card details already exposed.

Sansec also recommends clearing the cache before another order. Its warning allows up to 24 hours of continued malicious-script activity after cleanup, but the report supplies no precise cleanup time from which a shopper can calculate when that window ends.

Open the page → · show less

Eight of 14 recent listings fell on debut, while Slender West Lake postponed its October 23 offering timetable, citing factors including market conditions.

Jiangsu Slender West Lake Culture and Tourism postponed its global offering and planned Hong Kong Main Board listing two days before its scheduled October 23 debut. Its October 21 announcement cited several factors, including prevailing market conditions, according to Yicai.

The decision came alongside losses for buyers of recent offerings. Eight of the 14 stocks listed in Hong Kong since September fell below their offer price on their first trading day, Yicai reported on October 27, citing Yu Fenghui, a consultant at the Top 100 Hong Kong Listed Companies Research Center. Medcaptain Medical Technology fell nearly 43 per cent.

Hong Kong’s fundraising figures still show a record year through September. The exchange recorded 116 new listings that raised more than HK$388 billion in the nine months to September 30, more than twice the year-earlier proceeds and the highest total for any nine-month period in its history.

Ronald Wan, chief executive of Partners Capital International, told Yicai that mainland businesses seeking dual listings and advanced technology firms had driven the fundraising. He expects full-year proceeds of HK$480 billion to HK$500 billion, even as a gap has emerged in new IPO filings and some offerings have been delayed or abandoned.

For investors considering the next offer, the eight first-day declines are evidence of recent debut risk. They do not explain why each stock fell or predict the next listing’s performance. Nor does the record fundraising total establish how readily the next issuer can complete an offering.

Slender West Lake said it remained committed to completing its listing. But after considering factors including prevailing market conditions, it put back the global offering and the October 23 listing that investors had been due to assess.

Open the page → · show less

The order blocks a restriction that effectively forced lobbying firms to choose between Alibaba and clients seeking Pentagon contracts.

Washington lobbying firms driven away from Alibaba now have a preliminary court order blocking the restriction that forced a choice between clients.

The US restriction effectively required firms to choose between representing the Chinese technology company and representing clients seeking Pentagon contracts. Taking the Alibaba account meant choosing which clients to keep.

US District Judge Eumi K. Lee ruled on Monday that the government had failed to provide sufficient evidence to justify the restriction.

She found Alibaba likely to succeed on its free-speech challenge. That finding supports preliminary relief; it does not settle the challenge.

Alibaba’s Chinese military designation remains intact. The company has secured relief over who can represent it in Washington while retaining the designation that makes the account politically difficult.

For a lobbying firm considering the account, the preliminary order removes the restriction forcing a choice between Alibaba and clients seeking Pentagon contracts.

Open the page → · show less

A web-address design made shared conversations searchable, exposing a security problem that model testing alone cannot resolve.

Meta put the titles of shared Muse conversations in a part of the web address that outside services record. Researchers at security firm Claroc found 6,800 sharing links with a single query, according to Ars Technica.

Among the conversations they opened were an internal revenue review marked for a chief executive, a small-business loan application and a workplace vacation calendar. People had shared these pages by link. The design allowed strangers to find the links themselves.

The mechanism is straightforward. Before a browser opens a page, it asks the internet’s address system, DNS, where to find the site. Services called passive DNS databases collect records of those lookups. Muse placed each conversation’s title and identifying text in the subdomain, the part of the address those databases can capture. Once an intended recipient opened a link, its address could enter a searchable record.

A long, hard-to-guess address protects a page only while the address stays hard to obtain. Muse gave it a route into a directory.

That matters because Meta designed Muse to fill forms, book appointments and make purchases. Those tasks require access to users’ accounts and bring sensitive material into its work. Sharing an output then becomes another place where that material needs protection. The model can complete the task correctly while the surrounding software exposes the result.

Elsewhere this week, Latent Space reported researchers using GPT-5.6 Sol and Gemini 4 Argon to help discover and investigate PHP2Shell, vulnerabilities in PHP, software that runs websites. Researchers reported working exploits; maintainers and security providers coordinated fixes and temporary protections. That is evidence of useful security work beyond a benchmark score.

The two reports expose the distance between finding software flaws with AI and securing a product that uses it. A model-assisted investigation can uncover a defect deep inside PHP. It does not establish that a separate agent’s sharing system keeps confidential work confidential. Each path by which information leaves the product still needs its own examination.

For the person whose revenue review became discoverable, the failure occurred after the conversation was ready to share. Meta told Ars it had suspended that feature while reviewing the report and considering changes. Muse’s route from private work to a shareable page is now closed during that review.

Open the page → · show less

A harder horizontal bar routine carried Tang past an imperfect landing and turned Taiwan’s collection of world medals into a collection with a champion.

Tang Chia-hung left the landing untidy and Taiwan’s gymnastics history considerably improved.

On Sunday, October 25, in Rotterdam, he won the men’s horizontal bar final at the Artistic Gymnastics World Championships. His 14.966 points beat Japan’s Shinnosuke Oka and Sergei Naidin, who both scored 14.800. Oka took silver on the execution tie-break.

Taiwan had won world silver and bronze before. Tang supplied the missing metal.

The landing gave him something to worry about while the arithmetic went to work. Compared with qualifying, he had raised his routine’s difficulty by six tenths of a point. His execution score fell by two tenths. The harder routine covered the deterioration and left him four tenths better off overall. Tang had given himself enough room for an imperfect ending.

That was a consequential wager. Each finalist started from zero, with one routine to settle the title. Second place in qualifying bought Tang admission and precisely no insurance.

The rules reward difficulty, but offer no refund for failing to complete it. Tang’s 6.6 difficulty matched the highest in the final. Colombia’s Angel Barajas attempted the same difficulty and finished seventh. Ambition still needed delivery.

Tang arrived after finishing fifth on horizontal bar at the Aichi-Nagoya Asian Games. According to CNA, he reviewed that performance with coach Huang Bo-rui and a sports psychologist, then approached the world final as a practice session to reduce the pressure.

The result gives his next pursuit a different starting point. Tang told CNA he felt more confident about seeking qualification for the Los Angeles Olympics at next year’s world championships. Taiwan’s first world champion will still have to earn his Olympic place.

Open the page → · show less

Published on 23 October, the experiment demonstrates repeated error correction without measuring error signals or feeding back instructions.

The researchers repair a quantum memory by cooling its neighbour. In the University of Science and Technology of China’s experiment, the correction cycle works without repeatedly measuring an error signal and sending back instructions. Researchers working on trapped-ion memories get a demonstrated alternative to that control machinery, one that keeps the information usable longer than the system’s best unprotected qubit.

The Nature Physics study stores one logical qubit, a protected unit of quantum information, in several internal states of a single trapped ion. USTC’s account specifies four states and a second, helper ion. The two ions share a vibrating motion. That shared movement gives the researchers somewhere to send the disturbance before removing it.

USTC’s experiment principally corrects phase errors caused by magnetic-field fluctuations. Its multifrequency laser control moves the disturbed state back into the arrangement used to store the information. During that repair, it adds vibration to the ions’ shared motion. Components already in the correct arrangement remain unaffected. The repair leaves the disturbance in the movement.

Cooling empties that movement of its extra energy. As the Nature Physics study explains, the researchers cool the helper ion, which also cools the motion it shares with the memory ion. The memory retains the corrected information, and the shared motion is ready to receive another disturbance. Alternating repair and cooling makes the cycle repeatable without the error measurements and feedback that conventional approaches typically require.

USTC reports that, under ambient laboratory noise, the corrected memory’s coherence time, the measured duration for which it preserves quantum information, reached about 42 milliseconds. It outlasted the best physical qubit in the same system. That comparison is the experiment’s break-even test: the protected arrangement must preserve information longer than the best available unprotected one. Researchers get evidence that this correction cycle earns its place in a memory.

The publication dates put that advance in perspective. The team submitted its arXiv preprint on 23 April 2025; USTC dates the Nature Physics publication to 23 October 2026. This month brings journal publication of a previously public experimental result. The retrieved accounts establish those publication dates, rather than a date for the experiment.

The demonstrated stage is a trapped-ion memory experiment protecting one logical qubit against a particular kind of error. USTC identifies operations between multiple logical qubits, and protection during those operations, as further work. For researchers extending the scheme, the next gate is preserving the information while those protected units operate together.

Open the page → · show less

Intismeran added to pembrolizumab improved outcomes in 1,137 patients, giving tumour-specific immune training substantial clinical evidence.

The melanoma removed at surgery supplies the instructions for a treatment designed to attack what surgery may have left behind. Researchers analyse the tumour’s genetic changes, then build an mRNA vaccine encoding abnormal proteins specific to that patient’s cancer.

The removed tumour becomes a guide to recognising its remnants.

In INTerpath-001, a randomised, double-blind phase III trial involving 1,137 adults with completely resected stage IIB to IV cutaneous melanoma, adding intismeran autogene to pembrolizumab improved recurrence-free survival. Results presented at ESMO on 24 October put the two-year rate at 82.2%, compared with 70.0% for pembrolizumab alone.

That is a 12.2 percentage-point improvement in the estimated proportion alive without recurrence at two years. The two-year recurrence-free survival estimate starts at randomisation and counts patients who are alive without a recurrence.

The treatment begins with abnormal proteins called neoantigens, which provide distinctive targets for immune recognition. Researchers compare information from the patient’s tumour and normal DNA, then select targets associated with that cancer’s mutations.

Intismeran carries mRNA instructions for making those targets. The body’s cells translate the instructions into proteins and present them to the immune system, helping generate T cells that recognise cancer carrying the same targets.

Pembrolizumab works on another part of the process. It blocks PD-1, a receptor involved in restraining immune activity, allowing T cells to attack more readily. The vaccine supplies selected targets; the antibody releases an immune brake.

The developers’ headline figure, a 42% reduction, describes the relative hazard of recurrence or death during follow-up. It does not mean a 42 percentage-point improvement. The two-year rates make the absolute benefit easier to grasp.

The trial’s median follow-up was 24.6 months. ESMO describes overall-survival data as immature, with longer follow-up needed. The evidence establishes improved recurrence-free survival; a longer life or a cure has yet to be established.

Treatment also carries a burden. ESMO reported grade 3 or higher adverse events in 23.6% of patients receiving the combination, compared with 20.3% receiving pembrolizumab alone. Fever and injection-site pain were more common with intismeran. These broader adverse-event figures should be distinguished from the developers’ narrower figures for events attributed to treatment.

Merck Sharp & Dohme and Moderna funded the trial. Global principal investigator Georgina Long, of Melanoma Institute Australia and the University of Sydney, led research spanning 157 sites, including Australian centres. The developers say the results were accepted by the New England Journal of Medicine. The evidence supplied here comes from company, conference and university announcements.

The result applies to treatment after surgery for high-risk melanoma. It supplies no evidence that intismeran prevents melanoma in healthy people or works against other cancers.

For patients asking whether they can receive it, intismeran remains investigational. Availability will depend on regulatory approval in each country.

Open the page → · show less

The company that licenses Elvis is challenging whether AI training gets to use the catalog without buying permission.

Elvis comes with a licensing business. The permission slip is practically a backing vocalist.

A unit of Authentic Brands Group, which largely controls his intellectual property, joined other rights owners in suing Anthropic on October 23. Their allegation: the company behind Claude trained its models on 274 copyrighted songs without permission, including “Can’t Help Falling in Love” and “Don’t Be Cruel.”

That puts a specific commercial dispute underneath the famous face. These owners sell permission to use Elvis. They allege Anthropic used the compositions to build technology producing unlicensed derivative works. Anthropic did not immediately respond to Billboard’s request for comment.

Music’s appetite for an old catalog is hardly mysterious. Drake’s latest certifications made him the first artist past 350 million RIAA-certified units. The industry has excellent instruments for celebrating accumulated consumption. Permission requires separate paperwork.

AI companies argue that training qualifies as fair use, an issue still pending across dozens of lawsuits. That dispute reaches directly into Authentic’s licensing business: the owners are asking copyright law to give them control over this use of the songs. A successful fair-use defense would leave them unable to require a training license for the use challenged here.

Open the page → · show less