2026-11-19 Daily Digest
Technology News
- Anthropic lands on Nasdaq as ANT, record IPO sparks backlash ★ 9.0/10
- Anthropic IPO Prices at $83 Per Share on Nasdaq ★ 9.0/10
- Anthropic lists on Nasdaq, opens up 34%, market cap about $1.5 trillion ★ 9.0/10
- Anthropic Reportedly Debuts on Nasdaq at $2T Valuation ★ 8.0/10
- Anthropic says Claude helped solve two black hole physics problems ★ 8.0/10
- Ruby2ßhell: Pre-auth RCE reported in stock Rails ★ 8.0/10
- Microsoft agent-framework PyPI Package Hijacked to Install Malware ★ 8.0/10
- Google Releases Gemini 4 Neon With $3/$15 Pricing ★ 8.0/10
- PostgreSQL 19 Released with WAIT FOR LSN ★ 8.0/10
- OpenAI says its models cut chip area and lifted AMD GPU PPA by up to 14% ★ 8.0/10
- IEA warns data center carbon emissions will rise 70% by 2030 ★ 7.0/10
- European Commission deploys self-run Matrix tool to over 40,000 employees ★ 7.0/10
- Mid-latitude towns now see more satellites than stars overhead ★ 6.0/10
- US bankruptcy judge approves Google's $10M purchase of Spirit Airlines data ★ 6.0/10
- Casvisor 2.0 adds system accounts, database auditing, job center ★ 6.0/10
- MiniMax open-sources H3 Image for generation and editing ★ 6.0/10
- PHP to ship CRITICAL security fix tonight; Baota staffing engineers ★ 6.0/10
Technology News
Anthropic lands on Nasdaq as ANT, record IPO sparks backlash ★ 9.0/10
Anthropic listed on Nasdaq under the ticker ANT, issuing 980 million shares at $83 each for a public offering of about $81.34 billion, which reports say exceeds SpaceX's roughly $75 billion record from June this year. Nvidia, Microsoft, Google, TSMC, AMD, Broadcom and Micron also agreed to subscribe to an additional $26 billion of Class N shares at the offering price; after deducting about $1.96 billion in secondary share sales, Anthropic expects to receive about $105.4 billion. The prospectus disclosed $900.9 billion in compute and infrastructure purchase and lease commitments, of which $451.6 billion will be spent by the end of 2030, and listed "recursive" self-improvement as a core advantage. On the same day, nearly a hundred AI researchers, founders and scientists published the "Anti-Anthropic Statement," opposing the company's practice of warning about runaway risk while making accelerated AI development a selling point for its listing.
「Background」 Anthropic's IPO follows a rapid model-release cadence: the source says it launched the Claude Fable 5 flagship series in June, Opus 5.5 in September, and Fable 5.2 and Mythos 5.2 on November 9. Horizon's November 18 digest reported that OpenAI released GPT-6.2 Astra and the cheaper GPT-6.2 Sol, with OpenAI claiming Astra beat Claude Fable 5.2 on professional agent tasks, though developers criticized usability within 12 hours. The source describes that launch as a rushed response to Anthropic's release. External reports confirm the offering was priced at $83 per share for about 980 million shares, raising $81.3 billion, above an initially targeted $64.4 billion raise.
「Investor control and governance」 Public A-class shareholders and the seven chip and cloud companies buying N-class shares at the offering price will have no ability to challenge control: the prospectus assigns the T-class shares held by the Long-Term Benefit Trust the right to appoint a majority of the board, while the F-class shares held by the seven co-founders lock in 50.1% of total voting power. Investors therefore bear the financial risk of the roughly $900.9B in compute and infrastructure commitments disclosed in the filing without standard voting leverage over strategy.
References
Anthropic IPO Prices at $83 Per Share on Nasdaq ★ 9.0/10
Anthropic priced its initial public offering at $83 per share for 980 million Class A common shares. The company is selling about 956 million of those shares, while certain shareholders are selling roughly 23.65 million, with Anthropic stating it will not receive proceeds from the shareholder sales. Trading is expected to begin November 19, 2026, on the Nasdaq Global Select Market under the ticker ANT, with the offering expected to close November 20; underwriters also hold a 30-day option to buy up to an additional 147 million shares. The figures come from Anthropic's own announcement, and the item does not state a total amount raised or a valuation.
「Background」 Anthropic's IPO had been in the works for weeks before pricing: a November 10, 2026 roadshow announcement set the offering at 825 million shares with an expected price range of $70 to $78 per share, and ANT was then only a proposed Nasdaq symbol. The final terms announced here raise both the share count to 980 million Class A shares and the price to $83, above that earlier range.
「What It Means」 Because most of the offering consists of new shares sold by Anthropic rather than existing holders, the company stands to receive the bulk of the proceeds, while the shareholder portion does not benefit it. The 147 million-share underwriter option could increase the total number of shares sold if exercised within 30 days.
Anthropic lists on Nasdaq, opens up 34%, market cap about $1.5 trillion ★ 9.0/10
Anthropic (ticker ANT) listed on Nasdaq at an offering price of $83 per share and opened at $111.5, up 34% from the offering price. At the opening price, the company's market cap is about $1.5 trillion. The IPO issued 980 million shares, raising about $81.34 billion, of which the company issued 956.35 million shares and existing shareholders sold 23.6486 million shares. Anthropic also raised $26 billion through a concurrent private placement, with participants including strategic investors such as Google, Nvidia, Microsoft, AMD, Broadcom, Micron and TSMC, bringing the combined IPO and private placement to about $107.34 billion.
「Background」 Anthropic's listing follows a pricing process in which the IPO was set at $83 per share, a level that already valued the company at roughly $1.5 trillion and made it the largest public offering on record, surpassing SpaceX, according to AP reporting. Earlier coverage had described a reported price range of $70 to $78 per share and a fully diluted valuation of around $1.3 trillion, so the final pricing came in above that range before the 34% opening gain.
References
Anthropic Reportedly Debuts on Nasdaq at $2T Valuation ★ 8.0/10
Reuters reported that Anthropic made its Nasdaq debut at a $2 trillion valuation, an event that drew heavy debate on Hacker News. The headline figure is contested in the comments: one commenter said the stock fell after the first trade to roughly $1.26 trillion, and another cited a Yahoo Finance line calling it the worst first half hour for a big IPO since 2008. No source content was supplied, so the debut price, closing price, and share count could not be independently verified.
「Background」 Anthropic's listing follows a record IPO in which it raised $81 billion and priced shares at $83, with trading opening at $111.50, a 34% jump, according to press coverage of the debut. The company also planned a $26 billion private placement at the IPO price. Horizon's November 18 digest had reported OpenAI's launch of GPT-6.2 Astra and the cheaper Sol, with OpenAI claiming Astra beat Claude Fable 5.2 on professional agent tasks at 5% of the cost for equivalent work, framing the competitive backdrop against which Anthropic went public.
「Impact」 Commenters tied the listing to two concrete concerns: a weaker job market for software engineers and rising electricity demand from data centers. One commenter argued that a valuation below OpenAI's private-market target would put pressure on OpenAI's own fundraising picture. These are commenters' interpretations, not confirmed outcomes.
「Community Discussion」 Reaction split between those calling the listing a historic moment and those calling it a capital misallocation. An SF resident described fearing displacement and job losses, a software engineer said the "gravy train is basically over," and others questioned whether the $2T figure would hold given the reported drop to about $1.26T.
References
Anthropic says Claude helped solve two black hole physics problems ★ 8.0/10
Anthropic published a research post claiming Claude helped solve two open problems in black hole physics, according to a Hacker News submission. The available evidence is limited to the vendor's blog post and the accompanying papers; no independent verification or methodological detail is included in the supplied material. Commenters noted that the word "prompt" does not appear in the blog post, either abstract, or either paper, leaving unclear how much of the work was Claude's. One commenter described the first result as a complex, error-prone calculation and the second as an upper limit on black hole mass consistent with a given central charge.
「Background」 Anthropic's claim concerns Claude Fable 5.2, which the company says solved two open problems in theoretical physics, both related to black holes and the effort to reconcile relativity with quantum mechanics. The result was also listed as a bullet point in the Fable 5.2 and Mythos 5.2 launch announcement, according to a Hacker News commenter. The current item is a research blog post with no supplied source content, so the specific methods and prompts behind the claimed solutions are not available here.
「What to watch」 Because the claim rests on an Anthropic blog post and papers that, per commenters, omit the prompting details, readers evaluating the result should look for the specific calculations and any independent check before treating it as a solved problem. The framing also matters for how AI-for-science contributions get credited in future announcements.
「Community reaction」 Commenters pushed back on the "Claude solves" framing as anthropomorphizing, with one arguing it is like saying "it broke out of the sandbox" instead of "we failed to build an adequate sandbox." Another noted that many of the roughly 80 comments at the time were not about the papers at all, while one reader found the bootstrap paper hard to read but the blog post more accessible and human-written.
References
Ruby2ßhell: Pre-auth RCE reported in stock Rails ★ 8.0/10
A write-up published at verialabs.com describes "Ruby2ßhell," a pre-authentication remote code execution vulnerability in stock Rails that reportedly chains a single-byte overflow into full RCE. The supplied material provides no affected Rails versions, patch status, or independent reproduction, so the claim rests on the write-up itself rather than verified measurement. Commenters note the issue overlaps with an existing Ruby advisory that covers six other vulnerabilities, so it is not a wholly separate disclosure.
「Context」 The Ruby project published an advisory on November 18 covering seven vulnerabilities in its regular-expression engine and string case conversion, with proof-of-concept code showing some could trigger remote code execution; fixes shipped in Ruby 4.0.10, 3.4.13 and 3.3.14. Commenters on the Rails write-up note that it overlaps with that advisory rather than describing a wholly separate flaw. The disclosure also lands amid a broader wave of pre-authentication RCE reports, including Horizon's November 18 digest entries on Java4Shell in OpenJDK's Base64 decoder and the NGINX2Root chain.
「Impact」 If the chain holds, operators of unpatched stock Rails deployments face pre-authentication code execution, which is more severe than a post-auth bug because no credentials are needed. Readers should check the referenced Ruby advisory and their Rails version before treating this as a distinct new issue, since commenters say it is already covered there.
「Community discussion」 Commenters debated whether AI-assisted tooling is turning many single-byte overflows into reported RCEs by automating the tedious chain-building, with cowlevel arguing that if such overflows can be escalated with minimal human effort, the whole stack must be proven memory safe. barbegal said a single-byte overflow to RCE is achievable but sounds tough, while circularfoyers and djha-skin questioned whether this is new or just another entry in a fast-growing wave of critical CVEs.
References
Microsoft agent-framework PyPI Package Hijacked to Install Malware ★ 8.0/10
Microsoft's agent-framework package on PyPI, downloaded nearly 200,000 times a month and used to build and deploy AI agents, was hijacked to install information-stealing malware. The affected versions are v1.26.1 and v1.26.2, which steal environment variables, credential files, and other sensitive data for services including Azure, Git, Docker, and Hugging Face. The malicious code runs whenever the Python interpreter that installed the package starts, even if the package is never imported. A Microsoft spokesperson said the company set some package projects to private during its investigation.
「Background」 The malicious releases used `.pth` startup hooks, a Python mechanism that executes code whenever the interpreter starts, which explains why the payload ran without the package being imported. Socket's research team reported that the compromised versions targeted Azure, GitHub Actions, and CI/CD credentials, and a separate account states the two releases were live on PyPI for about 45 minutes on November 17, 2026. This incident follows another supply-chain compromise reported in Horizon's November 18 digest, in which Optimizely's Web Experimentation product was tampered with to plant malware on visitors of at least 178 customer websites.
「Impact」 Developers who installed v1.26.1 or v1.26.2 should treat credentials for Azure, Git, Docker, Hugging Face, and any other secrets exposed to that Python environment as compromised and rotate them. Because the payload triggers on interpreter startup rather than on import, simply avoiding use of the package does not prevent execution.
References
Google Releases Gemini 4 Neon With $3/$15 Pricing ★ 8.0/10
Google released Gemini 4 Neon, a cost-focused flagship model positioned between the lighter Hydrogen and the heavier Carbon in the Gemini 4 family. It carries a 1M-token context window, scores 61 on the Artificial Analysis index, and is priced at $3 per million input tokens ($0.15 for cached input) and $15 per million output tokens. Google says it is available immediately in Google AI Studio, the Gemini App, the enterprise API, and the Google Antigravity coding environment. The benchmark and pricing figures come from Google and the article's reporting and have not been independently verified.
「Background」 Gemini 4 Neon is the fourth model in Google's Gemini 4 family, following Hydrogen, Argon, and Carbon, and arrives roughly three weeks after Carbon, which the article describes as the model aimed at long-horizon, complex tasks. Google's own positioning places Neon between the cheaper Hydrogen (54 on the Artificial Analysis index) and the higher-priced Carbon (60), with Neon reaching 61 because it exposes an additional "xhigh" reasoning level. Artificial Analysis reported on November 18, 2026 that Neon's xhigh configuration scored 61 on its Intelligence Index, one point above Carbon at its high setting, at a lower cost per task.
「Cost impact for developers」 For developers choosing between frontier models, the article's cited comparison puts a single Neon task at $3.50 versus $7.20 for Claude Opus 5.5 at the highest reasoning setting, with Neon taking 20 minutes against Opus's 18. Because Neon is available immediately in Google AI Studio, the Gemini App, the enterprise API, and Google Antigravity, teams already on Google's stack can switch without a waitlist, though the benchmark and pricing figures come from the article and vendor materials rather than independent measurement.
「Community Discussion」 No substantive community comments were available for this item.
References
PostgreSQL 19 Released with WAIT FOR LSN ★ 8.0/10
PostgreSQL 19 has been released, roughly two months later than the usual September cycle; Simon Willison attributes the delay mainly to issues the team had to address amid ongoing AI-reported vulnerabilities. The release adds a WAIT FOR LSN command that lets a session on a standby wait until the standby has replayed the write-ahead log up to a specified log sequence number before running a query, providing read-your-writes consistency for applications that send reads to replicas. An optional timeout causes an error if the replica does not catch up in time. Willison notes the pattern resembles Wikipedia's hand-rolled MySQL replication-lag handling.
「Context」 PostgreSQL 19's general availability was pushed to November 19, later than the project's usual September release, as reported coverage attributed the delay to a surge of AI/LLM-reported bugs that the committers had to triage and fix. The release notes list other additions alongside WAIT FOR LSN, including a new REPACK command for reclaiming disk space and reorganizing table contents. The read-your-writes pattern itself is not new: MediaWiki has long implemented a hand-rolled equivalent against MySQL replication, saving the primary's binlog position and waiting for a replica to catch up before reading.
「Impact」 Applications that route reads to standbys can now capture the primary's LSN after a write (for example via pg_current_wal_insert_lsn()) and pass it to WAIT FOR LSN on the replica, rather than relying on application-level lag checks. The timeout option gives those applications an explicit failure mode when a replica falls behind.
References
OpenAI says its models cut chip area and lifted AMD GPU PPA by up to 14% ★ 8.0/10
OpenAI says its models contributed to the design of its in-house Jalapeño inference chip, reducing standard-cell area by about 13% in vector units and about 10% in matrix units, with the optimizations validated during chip bring-up. In work with AMD, OpenAI reports power, performance, and area (PPA) gains of up to 14% on flagship GPU IP relative to a highly optimized baseline, and says some changes have entered designs sent for manufacturing. OpenAI also announced a collaboration with Synopsys to train a GPT-Synopsys model. The figures come from OpenAI's own announcement and have not been independently verified.
「Background」 OpenAI and Synopsys announced GPT-Synopsys on September 30, 2026, a specialized model built to reason about chip design and verification and to operate Synopsys EDA tools directly, under a multi-year agreement with revenue sharing and joint go-to-market plans. The current announcement extends that collaboration and reports results from applying OpenAI models, including GPT-6 Astra, to circuit optimization in Jalapeño silicon and RISC-V core designs.
「Impact」 According to OpenAI's announcement, AMD also verified single-digit percentage PPA gains across key blocks of its upcoming next-generation flagship CPU using model-written designs, extending the claimed benefit beyond the GPU work. Because these figures come from OpenAI's own account and the source summary provides no independent verification, teams evaluating AI-assisted design flows should treat the reported gains as vendor claims pending external validation.
References
IEA warns data center carbon emissions will rise 70% by 2030 ★ 7.0/10
The International Energy Agency (IEA) released a report warning that data center electricity consumption will double by 2030 and carbon emissions are expected to rise by about 70%. The report says AI data centers will consume an additional 350 TWh of electricity each year over the next four years, more than the UK's total annual electricity consumption. The IEA expects that by the end of this decade, about half of the power for China's AI data centers will come from coal-fired plants, with the rest from renewables, a small share from nuclear and a very low share from natural gas; about 60% of US data center power will rely on natural gas. As of August this year, about 180 GW of new natural gas generation projects in the US are planned to supply data centers, almost equivalent to the EU's total existing installed natural gas generation capacity.
「Background」 The IEA's Energy and AI work, including its April 2025 report, already examined how data centre electricity demand affects energy supply, emissions and security. The current warning extends that analysis to 2030, adding country-level fuel-mix projections and planned US gas capacity.
"Impact" For data center operators and the utilities that power them, the report means electricity procurement over the next four years will rely more on fossil fuels, especially natural gas in the US and a coal-dominated power mix in China, which will directly push up their Scope 2 emissions and increase carbon compliance pressure.
References
European Commission deploys self-run Matrix tool to over 40,000 employees ★ 7.0/10
The European Commission completed a proof of concept in December 2025, confirming that the Matrix protocol and its Element client fully meet its needs, and then in September 2026 deployed a Matrix-based communication tool on every work laptop. The tool is operated by the European Commission itself and serves more than 40,000 users. Element's developer described the deployment experience in its blog. The timeline and scale above come from a brief Solidot report and provide no independent verification or technical details.
「Background」 Matrix is an open, decentralized messaging protocol, and Element is a Matrix-native client that supports features such as MatrixRTC-based voice and video calls. The European Commission's deployment followed a December 2025 proof of concept that tested whether Matrix and Element met its requirements, according to Element's account of the rollout.
"Impact" For organizations focused on secure communications, this case shows that self-hosted instant messaging based on open protocols can replace externally hosted solutions at a scale of tens of thousands of users; organizations intending to follow suit need to bear the server operations and client maintenance work themselves.
References
Mid-latitude towns now see more satellites than stars overhead ★ 6.0/10
An analysis by researchers at Accu Components, based on satellite trajectory data from October 12 and 18, found that 710 satellites pass over a typical UK town at any moment at night when the view is unobstructed, often exceeding the number of stars visible to the naked eye. Of those, about 105 are lit by the sun while the ground is dark and are bright enough to rival stars when crossing the zenith. The finding reflects the roughly 15,833 satellites now in low Earth orbit, most of them Starlink, whose orbits are inclined about 53 degrees to the equator and therefore concentrate over mid-latitudes; near the equator the average is 490. The figures come from a single small-company analysis with limited published methodology, not an independent measurement.
「Background」 Satellites become visible to the naked eye only when they are still sunlit while the ground below is dark, which is why sightings cluster around twilight and why not every satellite overhead can be seen. Starlink satellites orbit at an inclination of about 53 degrees to the equator, concentrating them over mid-latitude regions rather than the equator.
US bankruptcy judge approves Google's $10M purchase of Spirit Airlines data ★ 6.0/10
A US federal bankruptcy judge, Sean Lane, approved Google's $10 million purchase of internal data from the bankrupt airline Spirit Airlines. Spirit ceased operations in May, and Google won the data in an August auction as part of the asset liquidation; the data includes employee information such as emails, spreadsheets and calendar entries, as well as marketing, productivity and operational data. Google said it plans to use the data for product development and AI training, and last week agreed to additional privacy protections after unions and members of Congress raised concerns about the deal.
「Background」 Spirit Airlines ceased operations in May 2026, and its internal data was put up for sale as part of the bankruptcy asset-liquidation process; Google won the auction in August. Before the sale, a privacy ombudsman recommended in October that Judge Sean Lane approve the deal after passenger personal information was excluded from the roughly 97 million records involved, according to Reuters. The approval on November 17, 2026 followed that review and Google's agreement last week to add privacy safeguards for the employee and operational data.
「Impact」 The added safeguards mean the employee data Google acquired is subject to privacy conditions beyond the original auction terms, a constraint that follows scrutiny from unions and lawmakers over the sale of workers' internal communications and records.
References
- US bankruptcy judge approves Spirit Airlines data sale to Google - Reuters
- Google won the bankruptcy auction for Spirit Airlines' internal business data ...
Casvisor 2.0 adds system accounts, database auditing, job center ★ 6.0/10
Casvisor, an Apache-2.0 open-source bastion host, released version 2.0.0, adding multiple system accounts per asset, automated and on-demand password rotation, and audited web SQL consoles for MySQL and PostgreSQL. Permissions now carry an account field (asset account, same-name account, all accounts, or a named user), and the SSH gateway encodes the account in the username, for example `ssh -p 2222 alice@deploy@web-01@casvisor.example.com`; passwords stay server-side and permissions are rechecked when a session starts and when the WebSocket opens. The release also adds a job center that runs commands, Bash/PowerShell scripts, or file distribution (up to 1 GB) across assets selected by name or tag, plus an access-request workflow where users request time-bounded access and admins approve, narrow, or revoke it. The author states that password rotation works only on SSH assets, that port-forwarded database clients are not SQL-parsed, and that the job center has no Ansible playbooks or scheduled jobs.
「Background」 Casvisor is an Apache-2.0 open-source bastion host and security log auditing tool from the Casbin project, which also maintains the Casdoor identity platform. The author's earlier post on V2EX introduced the project and listed areas where it lagged behind JumpServer; version 2.0.0 is presented as closing most of those gaps. The release keeps the same deployment model, with data stored in the database and upgrades done by swapping the Docker image.
「Upgrade and compatibility」 Upgrading is a matter of pulling `casbin/casvisor:2.0.0` and restarting, since data lives in the database; existing permissions without an account field keep their prior behavior. Password rotation requires the operating account to have root or passwordless sudo on Linux, or administrator rights on Windows, and revoking, disabling, or deleting an account immediately disconnects active sessions.
References
MiniMax open-sources H3 Image for generation and editing ★ 6.0/10
MiniMax released the H3 Image model on November 11 with open weights available on Hugging Face and GitHub. The 33.1B-parameter model handles both image generation and editing, accepting up to 16 reference images addressed in prompts as Image 1, Image 2 and so on, and it supports output up to roughly 16 megapixels at 4K, including 6272 × 2688 at 21:9. Because it builds on the H3 video model, generated images can serve as a first frame for H3 Max video generation. Local use is impractical for most: the weights are 69 GB plus a 67 GB text encoder, and MiniMax lists H200 and RTX PRO 6000-class hardware. The license also excludes the US, EU, UK and South Korea from royalty-free use, so commercial users should check the terms.
「Background」 MiniMax's H3 Image is part of the same H3 family as the company's H3 video model, which Hugging Face lists as a 33B-parameter image-text-to-video model. The H3 Image repository describes a single checkpoint that handles both text-to-image generation and multi-image editing, producing roughly 16 MP at every aspect ratio in six sampling steps without an upscaler. The source post says H3 Image was built on the H3 video model so generated images can serve as a first frame for H3 Max video output.
「Impact」 Teams planning commercial deployment must verify licensing before building on the weights, since the royalty-free scope excludes several major markets. The hardware requirements also mean most users will rely on hosted inference rather than local runs; the author reports roughly 15–19 seconds per 2K image, about 22 seconds for 4K, and about 20 seconds for an edit with one reference image on a third-party site.
References
PHP to ship CRITICAL security fix tonight; Baota staffing engineers ★ 6.0/10
The PHP security team announced on November 18 that it plans to release a security-fix version between 22:00 and 22:30 Beijing time on November 19. The update, part of PHP's regular release cycle, addresses vulnerabilities affecting all supported versions, with the highest severity rated CRITICAL. PHP says the CRITICAL issue requires non-default configurations or uncommon runtime environments to trigger, and the specific vulnerability details and fixed version numbers have not yet been disclosed. Baota (BT Panel) said it has assigned engineers to stay on duty tonight to build, compatibility-test, and verify the new PHP release as soon as it is published, then push a panel update once testing passes.
「Background」 PHP's security team announced on November 18 that it would ship a security-fix release in the 22:00–22:30 Beijing time window on November 19, part of its regular release cycle, covering all supported branches. Baota's notice came before any CVE identifiers or fixed version numbers were published, so the specific flaws and patched versions were still unknown at the time of the post. Baota said it had engineers on duty to build, compatibility-test, and validate the new PHP versions and then push a panel update once testing passed.
「What to watch」 Operators running PHP through Baota should expect the panel update to lag the upstream release by the time needed for building and compatibility testing, so they should not assume same-night availability. Because the CRITICAL flaw is described as requiring non-default configurations, exposure depends on each deployment's setup, and the affected version range will remain unknown until PHP publishes the advisory and fixed version numbers.