Weekly AI Roundup: What Happened in AI This Past Week (August 15–21, 2026)

Welcome back. Late August in Dhaka has the tired quality of a long monsoon’s end — the rain still arrives, but nobody rearranges their day around it any more. There is a lesson in that for reading AI news. The announcements keep coming at the pace they did in April; what changes is how much of your attention each one deserves.

This week deserved more than most, because two of the biggest stories landed within twenty-four hours of each other and pointed in opposite directions. On Monday, OpenAI committed to eight gigawatts of computing capacity on a decommissioned Cold War site in rural Ohio. On Tuesday, the Governor of Pennsylvania signed an order saying that in his state, nobody builds anything like it without the neighbours agreeing first. Accelerator and brake, four hundred kilometres apart. Five global, five from home, and the same two questions on each — who benefits, who pays.

5 Big International Developments

1. OpenAI leased eight gigawatts in rural Ohio, and Nvidia guaranteed up to $105 billion of it

On August 17, OpenAI announced it had agreed to secure roughly 8 gigawatts-IT of capacity at the PORTS-Pike Technology Campus in Pike County, Ohio, on private land plus remediated federal land at the former Portsmouth Gaseous Diffusion Plant — a Cold War uranium-enrichment complex. SB Energy will build, own and operate the campus and lease it to OpenAI for twenty years; the site will exclusively host Nvidia compute. Nvidia is investing $1.5 billion in SB Energy and providing credit support for the first 4.25 gigawatts. In a filing with the US Securities and Exchange Commission the same day, Nvidia described that support as residual value guaranties capped at up to $105 billion, triggered only if OpenAI becomes insolvent or fails to pay, and beginning only after the facilities are ready for service, expected in 2028. OpenAI put the buildout at 35,000 construction jobs through 2032 and 2,500 permanent operating roles, with first power of 800 megawatts in 2028, and committed $40 million in community grants on top of SB Energy’s $40 million, plus up to $84 million in Codex credits at $100 each for roughly 844,000 eligible Ohio students.

Precision matters here, because several outlets blurred it. The $105 billion is not money Nvidia is spending; it is a ceiling on a contingent backstop, a promise to cover the landlord’s losses if the tenant defaults. (A plain-language note: “gigawatts-IT” means power delivered to the computers themselves, excluding cooling overhead. A “residual value guaranty” is an insurance-like promise that makes a lender comfortable enough to finance a building.) Jensen Huang addressed the obvious objection directly on X: “Is this circular financing? No. OpenAI will pay the lease.”

Who benefits, and who pays? OpenAI gets eight gigawatts without carrying the construction debt. Pike County — one of Ohio’s poorer counties — gets jobs and $164 million in stated benefits. The exposure sits with Nvidia’s shareholders in the default case, and with the region’s energy future in the ordinary case: Axios reports that 9.2 gigawatts of new gas-fired generation is ultimately envisioned. OpenAI’s own commitment is that SB Energy pays the full cost of grid upgrades and that “those costs will not be shifted to Ohio or other ratepayers in the region.”

The Dhaka view is to notice the sentence about ratepayers, because it is the sentence Bangladesh will eventually need. An AI data centre is already described as under construction at Kaliakair, and nobody has published its megawatts, its tariff, or who absorbs the transmission cost. Ohio got that in writing before the first slab was poured. Asking for the same is not obstruction; it is the ordinary price of hosting.

2. OpenAI paused its largest frontier training run over its own cyber evaluations

On August 18, OpenAI published a post saying it had “temporarily slowed the pace of scaling,” including a two-week pause in reinforcement-learning training on its latest models intended for deployment, and that “our largest planned frontier RL run remains on hold.” Two things triggered it: a security incident involving model evaluation on Hugging Face, and, in the company’s words, “preliminary evidence that one of our upcoming models, Astra, may meet the Critical cybersecurity capability threshold under our Preparedness Framework.” That internal determination is dated August 7. The company also paused frontier model inference in research clusters for any run that could execute code or use internet-capable tools, and says “a significant number of workloads remain paused” pending migration to isolated environments.

The monitoring system is the interesting part. Activation classifiers now run at every sampled token, escalating to higher-compute automated investigators, with a target of raising an alert within thirty minutes; if the safety, security and research teams cannot rule out a false positive inside that window, they are expected to pause the activity. “Once we determined that Astra may have critical cyber capabilities on August 7,” the post says, monitoring was extended to all inference of Astra with tools, not merely training. And the company put a price on it: “Our current estimates put monitoring overhead at roughly 20% of the inference compute being monitored.” OpenAI told The Register this is internal research spend that will not be passed to customers.

Who benefits, and who pays? In the short run OpenAI pays — twenty per cent of monitored compute is a real tax on a business whose whole constraint is compute. Everyone else benefits from the precedent: this is the first time a frontier lab has publicly halted its biggest training run because its own evaluations said stop, rather than because a regulator did. (A plain-language note: reinforcement learning here means training a model by rewarding good outcomes on tasks; “Critical cybersecurity capability” is OpenAI’s own top severity tier, meaning a model could meaningfully uplift someone attacking computer systems.)

The Dhaka view is that self-regulation of this kind is welcome and insufficient in the same breath. Welcome, because a company slowing down when its own instruments blink is exactly the behaviour we want. Insufficient, because the instruments, the thresholds and the decision all sit inside one company, and the countries that will absorb the consequences — including ours, with a public digital estate expanding faster than its security budget — have no way to audit the reading.

3. Anthropic says Claude autonomously designed working protein binders against 14 of 15 targets

On August 18, Anthropic published research reporting that its models, given a roughly 30,000-token seed prompt and then left to run without further scientific guidance, designed protein binders that succeeded against 14 of 15 targets — 354 confirmed binders from 1,320 designs, validated in wet labs by Adaptyv Bio and Twist Bioscience. For scale, Anthropic notes that the two largest public collections together hold about 770 binders from 5,700 designs against 40 targets. Reported hit rates were 26.7 per cent for its Mythos Preview model and 22.6 per cent for Opus 4.8 in multi-target mode, and 35.1 per cent in single-target mode, against a stated field baseline of 10–15 per cent. Against one target, RBX1, the single-target run hit 40 per cent versus 3.7 per cent for participants in a human design competition, and its best design outperformed the winning entry among 245 submissions. The failures are disclosed too: 90 designs against maltose-binding protein produced no confirmed binder.

(A plain-language note: a protein binder is a small designed protein that sticks to a specific target protein — the building block of a great many drugs and diagnostics. A “hit rate” is the share of designs that actually stick in a lab, not in simulation.) Anthropic did not build a biology-specific model; it orchestrated existing open-source protein tools with a general-purpose one. The company was explicit about the two-sided nature of the result: “The uplift provided by the increasingly autonomous research capabilities of AI models will undoubtedly speed the development of human therapies and fundamental scientific discoveries. However, such capabilities are also dual-use: without robust safety measures, they could enable bad actors to perform dangerous research, such as the development of bioweapons.” Protein design remains unavailable for general access, and Anthropic says opening a vetted scientist access programme is among its highest priorities.

Who benefits, and who pays? Therapeutic discovery gets faster and cheaper for whoever holds the access key — currently a small number of labs in a small number of countries. Biosecurity pays the risk. The honest tension is that the same locked door that keeps the capability from bad actors also keeps it from a pharmacology department in Dhaka.

The Dhaka view is that this is where a national AI policy earns its keep or does not. Bangladesh has a large generics pharmaceutical industry built on manufacturing rather than discovery. Capabilities like this are the first plausible route from formulation to origination — but only for institutions that can meet a foreign lab’s vetting bar. Building that credibility, the ethics boards and the audited labs, is unglamorous work nobody announces at a summit. It is also the entire ticket.

4. Pennsylvania made local approval a legal precondition for data-centre permits

On August 18, Pennsylvania Governor Josh Shapiro signed Executive Order 2026-05, directing the state’s Department of Environmental Protection to review a data-centre permit application only if the developer has made a legally binding commitment to a set of standards and has already received all required local approvals. The binding mechanism is a Consent Order and Agreement with penalties attached; a template was published the same day. All AI data-centre projects were immediately removed from the state’s Permit Fast Track Program, non-disclosure agreements with data-centre projects were declared “not permissible,” and the Department of Revenue was directed to withhold the sales-and-use tax exemption on data-centre equipment from applicants who fail the standards. The standards require developers to pay the full cost of new generation, transmission and distribution without shifting it to households; to run documented community outreach; to hire and train local workers; and to meet strict water conservation rules. Operators must report annual energy and gas consumption, peak hourly usage and water use. The order also directs the state to work with the utility commission so that data centres lose electric service before other customers if the grid is stressed, and so that grid reliability costs fall on data centres rather than other ratepayers.

The numbers behind it explain the urgency. The state’s environment department counts over 100 data-centre projects in public databases, 58 that have engaged on permitting, 15 that have applied for at least one permit, and only five with all permits for a first phase. Shapiro was blunt: “My message to data center developers is clear: if you can’t agree to our strict requirements and get the community where you want to build to say ‘yes,’ you’re not going to have the Commonwealth’s support either. These are some of the biggest companies in the world — they can afford to be good neighbors, follow the rules, and do this right.” Patrick Cicero of the Pennsylvania Utility Law Project framed the point in household terms: the order “seeks to ensure that we do not have a system where the wealthiest companies in the world get economic windfalls while a grandmother in Pennsylvania gets stuck with an electric bill she cannot afford.”

Who benefits, and who pays? Households and municipalities gain leverage — particularly from the ban on non-disclosure agreements, the quiet radical clause, since NDAs are how these deals normally stay invisible until they are irreversible. Developers pay in slower permitting and a lost tax exemption, and critics note fairly that an executive order can be undone by the next governor.

The Dhaka view is to read stories one and four as a matched pair. On the same day, one jurisdiction promised to pay its own energy costs and another made that promise enforceable. When an investor arrives in Dhaka or Chattogram with a data-centre proposal and a request for confidentiality, the correct response is no longer improvised — it is a Pennsylvania-style list. Publish the megawatts. Publish the water. No NDA. Local consent first.

5. A Chinese humanoid maker opened 629 per cent above its IPO price

On August 19, Unitree Robotics — legally Yushu Technology — listed on the Shanghai STAR Market, becoming mainland China’s first publicly listed humanoid robot maker, and opened at 1,100 yuan against an IPO price of 150.80 yuan, a gain of 629 per cent. Caixin reported it as the largest first-day opening gain for a new STAR Market listing this year, valuing the company at roughly 444.9 billion yuan, about US$66 billion. The company raised 6.1 billion yuan, around US$905 million, against a 4.2 billion yuan target. It gave back some of that and closed at 845 yuan, still up about 460 per cent, near 342 billion yuan. Founder Wang Xingxing, 36, holds 68.8 per cent of voting rights. At the opening price, early backer Meituan’s 420 million yuan stake was worth 38.6 billion yuan.

Two things sat awkwardly beside the celebration. The day before, Reuters published an investigation tracing Unitree’s quadruped lineage to openly published research funded by the US Army from 2010 to 2020 involving MIT, the University of Pennsylvania, Boston Dynamics and NASA’s Jet Propulsion Laboratory. Ben Katz, formerly of MIT’s Biomimetic Robotics Lab, said the dimensions of Unitree’s Go series matched a robot he helped build, the Mini Cheetah, “to the millimeter”; Gavin Kenneally, formerly of the University of Pennsylvania, said the Army-funded work became “basically, the first Unitree robot that had any kind of scale.” Meanwhile, US regulators are moving to restrict imports of Unitree robots on security grounds.

Who benefits, and who pays? Wang and early investors capture an extraordinary windfall. Retail buyers who oversubscribed the offering now carry the valuation risk on a company whose largest potential export market may be closing. The broader beneficiary is the field: this is the clearest public-market signal yet that investors are pricing embodied AI — robots that act in the physical world — the way they priced language models in 2023.

The Dhaka view is about the research lineage, not the share price. American taxpayers funded quadruped locomotion work; the findings were published openly, as publicly funded science generally should be; a Chinese company built a global business on them. That is not theft, and calling it theft leads to the wrong conclusion — that openness was the mistake. For a country like Bangladesh, which cannot fund frontier robotics and depends entirely on reading and building on what others publish, open publication is not a loophole. It is the only door we have.

5 Stories That Hit Closer to Home (Bangladesh)

1. AI cameras on 490 pillars now issue fines automatically on the Dhaka–Chattogram highway

On August 18, Home Minister Salahuddin Ahmed inaugurated an AI-based automatic fine system at the Highway Police Command, Control and Monitoring Centre at Meghnaghat in Sonargaon, Narayanganj, with AI cameras installed on 490 pillars along the Dhaka–Chattogram highway. It had been announced two days earlier. It is the first such deployment on a highway; earlier AI camera systems operated within Dhaka city. It detects speeding, dangerous overtaking, disobeying signals, wrong-side driving and illegal parking. When a violation is detected, vehicle information goes directly to the BRTA database, an SMS goes to the registered owner’s mobile with the offence and the fine, and the owner pays from home through mobile financial services. Repeated violations can trigger cancellation of a vehicle’s registration under BRTA rules. The minister said the initiative would bring “a major change in highway traffic management by ensuring greater road safety, maintaining traffic discipline, reducing accidents and enabling swift, effective and transparent action against traffic law violations,” and that after evaluating this corridor, “the AI-based automated system will be introduced on all highways in phases.”

This is the largest operational AI deployment by the Bangladeshi state we have covered, and the ambition is right. The Dhaka–Chattogram corridor is the country’s freight artery and one of its deadliest roads, and automated enforcement removes the discretionary roadside stop, where much of both the danger and the petty extraction happens.

But the pipeline as described runs camera → vehicle registry → SMS → payment, with no human adjudication step mentioned in any account and no appeal procedure disclosed in any source we could find. That gap is not hypothetical: plate recognition fails on damaged, obscured or rain-blurred plates, and someone who receives a wrong fine has no described route to contest it. Two days later, on August 20, at a BIISS seminar in Dhaka titled “AI and Governance,” Md Ismail Zabiullah, Adviser to the Prime Minister’s Office, said Bangladesh needs caution in adopting AI over data privacy and consent and stressed finalising the national AI policy; BTRC Chairman Major General (retd) Md Emdad Ul Bari, chairing, called for a national interoperability framework, clear data-consent mechanisms and transparent AI procurement standards. Both were describing, in the abstract, the system that had gone live forty-eight hours earlier.

Who benefits, and who pays? Road safety on the busiest freight corridor benefits, genuinely. Mobile financial services providers gain a new stream of state collections. Drivers and vehicle owners pay — including the ones the system gets wrong, who currently have no published way to say so.

The Dhaka view is that the sequencing is backwards but recoverable. Deploy first and write the policy later is how most states actually do this. What is not optional is publishing, before the rollout to other highways, three things: the appeal procedure, the measured accuracy of the plate recognition, and who holds the footage and for how long. A system that fines citizens automatically owes them an automatic way to be heard.

2. Startup Bangladesh opened a Tk400 crore fund of funds — public money for VC firms, not startups

On August 16, Startup Bangladesh Limited, the state venture capital company under the ICT Division, began operations of the Bangladesh Fund of Funds with an initial size of Tk400 crore, roughly US$33 million, and issued a Request for Expression of Interest to professional venture capital fund managers. The structural change is the point: instead of investing directly in startups, the vehicle invests through selected local, regional and international fund managers, who must allocate at least a matching amount toward Bangladesh — a one-to-one requirement. Startup Bangladesh’s own release notes that Bangladeshi startups raised about US$1.2 billion over the past decade, of which local investors accounted for only about 7 per cent. Faqir Mahbub Anam, Minister for Posts, Telecommunications and Information Technology, said the company has already invested about Tk111 crore in 36 technology-based startups, and called the new vehicle “a game-changing platform for our startup ecosystem… the most important initiative that we have taken for the country to support our startups.” Rehan Asif Asad, Adviser to the Prime Minister for Posts, Telecommunications and Information Technology, added a caution worth keeping: “Not only financing, but startups also need mentorship, skills, market linkages and technological support.” The current fiscal year carries a Tk500 crore allocation for startup development, plus a zero per cent turnover tax and other exemptions.

(A plain-language note: a fund of funds does not pick companies. It picks the people who pick companies, and takes a small share of many portfolios instead of a large share of a few. The theory is that Tk400 crore, matched one-to-one, pulls in far more capital than it could deploy alone, and builds a local fund-management industry that outlasts the money.)

Who benefits, and who pays? Fund managers who win mandates benefit first and most concretely. Later-stage startups that currently hit a wall after seed funding benefit next, if the theory works. The taxpayer carries first-loss risk, and the honest failure mode is that Tk400 crore mostly funds management fees if the selection process is weak.

The Dhaka view, from an AI lab’s chair, is that this is the first plausible domestic capital route for AI-native companies — and that the selection criteria are the whole story. If they reward fund managers with real technical diligence capacity, Bangladesh gets investors who can tell a genuine model company from a wrapper with a landing page. If they reward relationships, we get a slower way to fund the same things. Publishing the criteria and the eventual scorecard would cost nothing and settle the question.

3. The spectrum argument opened, with November 2026 on the calendar

On August 15, at a Telecom and Technology Reporters’ Network Bangladesh roundtable in Dhaka, the government and the mobile operators began negotiating in public over the spectrum renewal cycle that starts in November 2026 and runs to 2034. Rehan Asif Asad said 79.4 megahertz of the 406 megahertz currently assigned to the four operators falls due for renewal in 2026 — about a fifth of the total — and repeatedly urged calm: “Please do not be tensed. This is only 20 percent and we’ll get it right InshAllah. What matters and what’s critical is the signal.” Robi’s Chief Corporate and Regulatory Officer Shahed Alam said global operators spend around 7 per cent of revenue on spectrum, against 15 to 18 per cent here. BTRC Chairman Major General (retd) Md Emdad Ul Bari put the supply picture plainly: Bangladesh has about 2.29 megahertz of spectrum per million subscribers, at the bottom of the 2-to-6 global range, while a 5G scenario needs 6 to 12 per million — implying a requirement of around 1,500 megahertz by 2035. He also noted that BTRC had asked the industry for two years to commission an in-depth market study, none was done, and the commission has now engaged its own team, adding that he doubts it can finish in the time available.

The handset half of the conversation was more concrete. Rehan Asif Asad said around half of mobile users are still on 2G feature phones, and that he had asked 14 local manufacturers to bring the cheapest smartphone from Tk10,000 down to Tk3,000; they came back with Tk5,000 to Tk6,000 as feasible, citing a National Board of Revenue commitment of zero tax and duty on raw materials for such devices. The chairman’s own framing of the trade-off was the sharpest line of the day: “I will not support services to save business; I will support business to save services.”

Who benefits, and who pays? Cheaper spectrum benefits the four operators immediately and, in theory, anyone building data-hungry products in Bangladesh. The treasury forgoes revenue. The risk, and it is the historical pattern, is that a price cut is banked rather than converted into coverage and quality.

The Dhaka view is that the renewal should be conditional, not merely cheaper. Tie relief to published, measurable service commitments — and the mechanism to verify them arrived four days later, which is the next story.

4. BTRC published five years of complaints: 66,705 since 2021, and monthly hearings from now on

On August 19, at its seventh public hearing, held online, BTRC disclosed that 66,705 customer complaints were filed against the four mobile operators between 2021 and mid-2026 — Robi 24,874, Grameenphone 21,526, Teletalk 11,594 and Banglalink 8,711. For the year from July 2025 to June 2026 alone there were 13,009 complaints, of which 11,279, or 86.70 per cent, were resolved; the commission said the figure covers complaints resolvable immediately and that it is not fully satisfied with the rate. Complaint categories included unauthorised value-added service activation and deductions, number portability, billing errors, SIM registration disputes, mobile financial services, fraud and deceptive prize promotions. Two structural commitments followed: BTRC will publish official monthly quality-of-service reports on its website from September, and Spectrum Division Commissioner Mahmud Hossain said the commission will hold a public hearing at least once every month under Section 87 of the Bangladesh Telecommunication Act, 2001 as amended in 2026.

Read alongside the roundtable four days earlier, this is a coherent regulatory sequence rather than two unrelated events. Operators argue they cannot invest because spectrum is expensive; consumers have five years of documented failures; the regulator’s answer is a monthly scorecard. The commission’s claims that a recent drive test found no operator below the quality threshold, and that tower radiation is within safety limits, are single-sourced to the regulator itself and unaudited — worth noting rather than repeating as established.

Who benefits, and who pays? Consumers benefit from a recurring public dataset, and so does anyone doing research on Bangladeshi connectivity — including us. Operators pay in scrutiny, arriving precisely as they lobby for cheaper spectrum.

The Dhaka view is that monthly published quality-of-service data is the most useful thing in this week’s Bangladesh news, and almost nobody will report it. Machine-readable QoS by operator and district is the kind of boring public dataset that lets researchers, journalists and startups build things nobody at BTRC has thought of. One request: publish it as CSV or JSON, not a PDF of a scanned table.

5. Airtel disappeared from Bangladesh after sixteen years

On August 17, at a launch in Dhaka, Robi Axiata PLC retired the Airtel consumer brand and replaced it with Cirkle, ending Airtel’s sixteen-year presence in Bangladesh. Airtel entered the market in 2010 and merged with Robi in 2016, but continued as a separate consumer brand with its own identity and the 016 numbering series. Existing customers become Cirkle customers automatically; 016 numbers, services, packages and balances are unchanged, as are mobile financial services accounts and wallets. The new brand is positioned as affordable and digital-first, with stated plans to introduce services using emerging technologies including artificial intelligence, aimed at entertainment, gaming, content creation and online communities. Managing Director and CEO Ziad Shatara said, “While the identity evolves, the brand promise remains the same: bringing people closer through trusted, affordable and always-on connectivity.”

On its own this is a marketing story. What makes it worth a slot is the timing. Robi collapsed a dual-brand structure it had paid to maintain for a decade, in the same week its sector was publicly arguing that costs are unsustainable — small but real evidence that the cost pressure operators describe is not purely rhetorical. And the AI mention is not decoration: a youth-focused mass-market brand built to ship AI-flavoured services is one of the very few AI distribution channels in Bangladesh reaching tens of millions of people who will never open a chatbot on a laptop.

Who benefits, and who pays? Robi benefits from a simpler cost base. Customers are held harmless in the immediate mechanics — same number, same balance. The country loses a distinct competitive identity in an already concentrated four-player market, a small cost paid slowly.

The Dhaka view is that whoever defines what “AI features” mean on a Tk5,000 handset will shape most Bangladeshis’ first encounter with this technology. If that is a chatbot bolted onto a data pack, the opportunity is wasted. If it is Bangla voice interfaces that work for someone who cannot read comfortably, it is the most consequential AI deployment in the country. Nobody is watching this closely enough.

Wrapping up

Ten stories in seven days, and the global five arranged themselves into an unusually clean argument. On Monday OpenAI committed to eight gigawatts in Pike County, Ohio, with Nvidia backstopping up to $105 billion of the lease. The next day, Pennsylvania made local consent and full cost-bearing binding conditions of a data-centre permit, and banned the non-disclosure agreements that normally keep such deals invisible. Also on Tuesday, OpenAI paused its largest frontier training run because its own evaluations suggested an upcoming model might cross a critical cyber threshold, at a monitoring cost of roughly 20 per cent of the compute being watched. Also on Tuesday, Anthropic reported that a general-purpose model, left to run on its own, produced 354 lab-confirmed protein binders and beat human competitors on a target where they managed 3.7 per cent. And on Wednesday, a humanoid robot maker opened 629 per cent up in Shanghai, on foundations laid by openly published American military-funded research.

At home, Bangladesh built things. AI cameras on 490 pillars now issue fines automatically along the Dhaka–Chattogram highway, with no published appeal route — while two days later, at a seminar across town, senior officials called for exactly the data-consent and procurement standards the new system lacks. Startup Bangladesh opened a Tk400 crore fund of funds and asked the market to match it one-to-one. The spectrum argument opened ahead of the November renewal, with the BTRC chairman noting the country holds 2.29 megahertz per million subscribers and will need around 1,500 by 2035. BTRC published 66,705 complaints against four operators since 2021 and promised monthly hearings and monthly quality data. And a sixteen-year-old brand became a new one with AI in its product plan.

One thread ties the week together, and it is the oldest one in this newsletter: capability is arriving faster than accountability, at every scale. OpenAI paused itself because no one else could. Pennsylvania wrote down what a data centre owes its neighbours because the market would not. Bangladesh switched on automated enforcement before writing the appeal procedure. In each case the technology worked; what lagged was the sentence explaining who answers when it goes wrong. That sentence is cheap to write and expensive to omit.

One outstanding item, for the record. The special legal panel on deepfakes and cyberbullying was announced on 21 July by Zahed Ur Rahman, the Prime Minister’s Adviser on the Ministry of Information and Broadcasting, who said the administrative work was complete and the formal notification would follow within a day or two. Thirty-one days later there is still no publicly reported gazette notification. We found no announcement this week either way, and we do not claim the panel does not exist — only that no public confirmation of its legal existence could be found. We will keep counting.

If something here caught your interest, share it with someone who does not follow tech. The conversation only spreads if we keep starting it.

See you next week. Stay well.