AI Weekly Malaysia

WhatsApp is rolling out parental controls for teen accounts, gated behind a parent-set PIN: guardians can restrict Channels use, who can see the teen's Status and profile photo, who can add them to groups, and get notified when a teen joins or leaves a group or when a group hits 30, 100, or 250 members. Parents can also set Meta AI content settings, choosing the default 13+ setting or a stricter 'Limited Content' option that disables private-processing features such as incognito chat and message summaries. WhatsApp says messages and calls remain end-to-end encrypted, and the piece notes Meta agreed in August to pay an $18 billion settlement with 29 U.S. states over child-safety claims.

Why: For most builders this is a consumer feature with no action item. It only bites if you ship on WhatsApp (Business/Cloud API) or run AI chat for under-18 users: Meta is treating private-processing AI features — incognito chat and message summaries — as the specific things to switch off for teens under a parent PIN, so an age-gated toggle for on-device or private-processing features is the pattern you'd likely be asked to match. The 30/100/250 group-size thresholds are the concrete notification triggers to note if you manage group-based messaging for minors.

Tom's Hardware flagged a Ugreen 10-port gigabit switch with PoE+ and a total power budget of up to 60W at under $38, described as a new record low price. Eight of the ten ports supply Power over Ethernet, aimed at cameras and Wi-Fi extenders. The body of the article is almost entirely site navigation and newsletter boilerplate, so no model number, PoE standard revision, per-port wattage, or switching capacity is given beyond the headline figures.

Why: Do the arithmetic before buying: 60W shared across eight PoE ports averages roughly 7.5W per port if everything is powered at once, which is fine for cheap IP cameras and Wi-Fi extenders but will not drive 802.3at devices that want up to 30W each. If you are speccing a small camera or AP rollout for an office or homelab, this price point is worth noting as a floor, but the missing specs here mean you should confirm the per-port budget and standard from the product page before ordering at volume.

Cloudflare says its Impact programs will cross $100 million in donated services during its 16th Birthday Week. Project Galileo, launched in 2014, now covers more than 3,500 domains across 120+ countries and blocked over 38.5 billion attacks in 2025 (about 105.4 million per day); the Athenian Project (2017) covers 440+ election websites in 33 U.S. states plus election bodies in eight countries, and Cloudflare for Campaigns (2020) runs with Defending Digital Campaigns.

Why: Nothing here changes what a Malaysian developer or founder ships this week — there is no product, price, region, or API change, and no Southeast Asian program is named. The one transferable detail is the origin story: Cloudflare's free tier started as a way to collect attack data from small sites and turn it into sellable products, and only later became a formal Impact commitment. If you run a free tier, that is the concrete decision to revisit — is free usage feeding your product roadmap, or is it just a cost centre you call community goodwill?

Samsung's Galaxy Tab S12 Ultra and S12+ go on sale in Malaysia on 7 October 2026, all in a single 12GB/256GB config, with the Ultra at RM5,999 (WiFi) / RM6,649 (5G) and the S12+ at RM4,999 / RM5,649. From 7 October to 8 November 2026 an instant rebate of RM500 (Ultra) or RM400 (S12+) applies when paying with eligible bank cards or SPayLater, dropping the S12+ WiFi to RM4,599, and Samsung bundles a Book Cover Keyboard plus S Pen accessories worth up to RM1,208 and subscriptions worth RM1,123.58 including 6 months of Google AI Pro and 4 months of Adobe Photoshop. Students and education staff can get up to 20% off via the Education Store (.edu email verification), with an extra 5% for first-time education purchases. Both run One UI 9 with MediaTek Dimensity 9500 and Dynamic AMOLED 2X, and are described as the first Android tablets to ship with Adobe Photoshop out of the box.

Why: This is a consumer hardware promo, not a platform change, so it only matters if you were already budgeting for a tablet: the education pricing (S12+ WiFi at RM3,999.20, Ultra WiFi at RM4,799.20, plus 5% first-time discount) and the 7 Oct–8 Nov rebate window are the two levers that change the price, and the rebate requires an eligible bank card or SPayLater. Note the article labels the S12+ price list under 'Galaxy Tab S12 Ultra' headings, so confirm the exact SKU and 5G/WiFi variant with the seller before ordering.

BAG Ventures, founded by former Google VP Bonita Stewart and former CapitalG partner Jackson Georges Jr., closed an $11.3M fund for early-stage AI startups after roughly two years of investing out of it. It has already backed 10 companies — including SXD, AI travel agent BizTrip, and agentic reasoning platform Nomadic — with check sizes of $100,000 to $500,000, and plans to deploy the rest over the next two years across AI infrastructure, compute, physical/edge AI, security, governance, and vertical SaaS. The pair describe their edge as access: warm introductions to enterprise operators rather than capital alone.

Why: If you are raising in the $100K–$500K band for an enterprise-facing AI product, this is one more pre-seed/seed option, and the pitch is explicitly about operator introductions into buyer organisations, not just money — so the useful question is whether that network maps to your target accounts. Nothing in the text ties the fund to Malaysia or Southeast Asia, so there is no stated local angle here; treat it as background on where early AI money is going, not as an opportunity you must act on.

Grab Malaysia launched a beta trial called Kid-Friendly Rides, assigning vehicles fitted with certified child restraint systems or booster seats so parents don't have to bring their own. For now it is only bookable through Advance Booking for airport rides to and from KLIA 1 and KLIA 2, with phased expansion planned across the Klang Valley, then Penang and Johor Bahru. Grab says enrolled driver-partners get guidance on installing, maintaining and handling the seats before pick-ups.

Why: If you build on-demand or marketplace products in Malaysia, this is a concrete example of launching a compliance-heavy SKU: child restraint systems have been mandatory here since 1 January 2020 and must meet UNR R44 or R129 for children under 36 kg, under 136 cm, or below 12 years old, and Grab gated the launch to one route type (KLIA advance bookings) while it trains supply. For anyone actually travelling with kids, the practical limit is that you cannot use it for a normal city ride yet — and there is no API, pricing detail, or developer-facing change in this announcement, so nothing to build against today. MIROS Director-General Siti Zaharah Binti Ishak is cited saying booster seats cut injury risk by 45% for children aged four to eight versus seat belts alone, which is the safety case the service rests on.

OpenAI announced a partnership with America's SBDC — a US network of small business development centers that says it serves 1 million entrepreneurs a year — to train roughly 150 advisors through its OpenAI Academy Community Trainer Program, with a stated goal of reaching at least 1,000 small businesses via in-person workshops plus one-on-one advising. Alongside it, OpenAI published a report, 'Small Businesses, Bigger Capabilities', claiming that about 4 million small-firm employees worldwide used OpenAI tools in the single week of September 9-15. The pilot combines an advisor training and credentialing pathway with a separate facilitator pathway and a three-hour workshop module.

Why: This is OpenAI's own announcement about its own program and its own usage report, and it is entirely US-based, so nothing here changes what a Malaysian builder ships this week. The one detail worth extracting is the delivery model: advisors and facilitators get credentialed first, then workshops scale. If you sell AI tooling or run training to Malaysian SMEs, that is a distribution pattern you can copy or compete with — and the closest local equivalents (SME Corp, MDEC, state-level SME digitalisation programs) are the channels to watch for a similar program appearing here.

A Tom's Hardware piece reports a former EVGA employee's account of the deteriorating relationship between EVGA and Nvidia ahead of the companies' 2022 split, citing Nvidia's Founders Edition pricing, pricing mandates imposed on partners, and forward-looking technology decisions as sources of friction. The excerpt supplied here contains only the headline and Tom's Hardware membership/subscription boilerplate — no quotes, figures, dates, or specifics from the actual reporting are present.

Why: There is no actionable detail in the supplied text: no pricing numbers, no margin figures, no named policies, no dates beyond the 2022 split referenced in the headline. If you build or buy GPU-dependent infrastructure, the only decision you can make from this excerpt is to go read the full article before repeating any of its claims — the headline alone does not establish what Nvidia actually mandated or what it cost partners.

Google's Mandiant Consulting and Threat Intelligence Group observed threat actors in September 2026 exploiting CVE-2026-88772, a CVSS 9.5 memory overflow in the DTLS record parsing of the NetScaler Packet Processing Engine (NSPPE) in Citrix NetScaler ADC and Gateway appliances. Malformed or fragmented DTLS record headers corrupt heap memory and divert control flow to shellcode with root privileges on the underlying FreeBSD platform, bypassing authentication entirely. Post-exploitation, attackers modify httpd.conf so .deb files are handled as PHP, stage web shells in /netscaler/gui/vpn/scripts/linux, and deploy WHIPSHOT (PHP web shell hiding Base64 C2 in native HTTP headers) plus SLAPSHOT (a Python tunneler proxying into internal networks for reconnaissance and credential theft).

Why: Only relevant if you, a client, or a vendor-managed environment actually runs NetScaler ADC or Gateway as an edge/VPN appliance: this is pre-auth root, so an unpatched box is a direct path to internal credential theft, and the httpd.conf change treating .deb as PHP plus shells under /netscaler/gui/vpn/scripts/linux are concrete detection artifacts to check. Everyone else has nothing to change here. Note the reported targeting is organizations in North America and Europe across government, financial services, technology, education, and legal sectors - the text gives no Malaysia or Southeast Asia angle.

Indosat Ooredoo Hutchison shareholders approved at an Extraordinary General Meeting on 29 September 2026 an extension of president director and CEO Vikram Sinha's tenure, first announced in July 2026, along with a change to its Board of Commissioners that accepted the resignation of commissioner Seppalga Ahmad. The company frames the continuity around its 'AI North Star' strategy, citing Zankore with 200MW of AI capacity secured through multi-year contracts, RAIA Grid backed by 86,000 kilometres of fibre, and an ambition to empower up to two million people through digital talent work by 2030.

Why: This is a governance and capacity announcement, not a product you can use today: there is no pricing, API, region, or availability date for Malaysian or SEA builders to act on. The one concrete thing to track is the 200MW of contracted AI capacity plus 86,000 km of fibre in Indonesia, which is a signal about where regional AI compute supply may grow — relevant if you are planning training or inference capacity outside Singapore, but not yet a reason to change any decision.

watchTowr published exploit details for CVE-2026-88772 (CVSS 9.5), a pre-auth memory overflow in Citrix NetScaler ADC and Gateway's DTLS handshake handling inside the NetScaler Packet Processing Engine (NSPPE), which CISA says is under active exploitation in the wild. The bug is a parsing inconsistency: the handshake header's length field declares a 120-byte message while each fragment's fragment_length field says 1 byte, so reassembly stitches roughly 174 KB of NetScaler Buffer data into a scratch buffer of only 35,840 bytes because the vulnerable version never checks whether the next packet fits. Individual packets are 1,459 bytes, and the flaw can lead to remote code execution or denial-of-service.

Why: If you or your employer don't run NetScaler ADC or Gateway, nothing in your stack changes this week — this is a patch-now item only for teams with that appliance in front of their services, since the path is pre-auth and exploitation is already observed. The transferable lesson is narrow and concrete: the overflow happens because the code trusted a declared per-fragment size (1 byte) while keeping the whole 1,459-byte record, so if you write any upload, websocket, or protocol reassembly handler, check whether you validate declared lengths against what you actually copy into a fixed buffer.

A Malay Mail Tech piece (published 30 Sep 2026) argues that spam has moved beyond email into internet-connected calendars, citing cybersecurity reports from Sublime that show a sharp increase in spam calendar invites, particularly affecting Google Workspace users, with predictions of continued growth. The article's stated remedies are reviewing calendar settings, limiting access, and consulting trusted cybersecurity experts. It names no figures, versions, affected organisations, or specific attack tooling.

Why: There is little here to act on: no incident counts, no named vulnerabilities, no Malaysia-specific detail, and the mitigation advice is generic. The one concrete claim is that calendar invites are now a spam/phishing channel aimed at Google Workspace users — so if your team runs Workspace, treat calendar invites as a second inbox and decide who can add events to shared calendars, rather than assuming email filtering covers you. Beyond that, this article does not justify a settings change on its own; wait for a report with numbers or a vendor advisory before reconfiguring anything.

Anthropic says Claude for Government is now generally available for US federal and state agencies, running in a FedRAMP High authorized environment after a public beta that started in July. Agencies get the same capabilities as commercial customers on the commercial release cadence, plus admin controls: no seat fees, prepaid usage in fixed increments with a hard not-to-exceed cap, department-level allocation of usage to sub-agencies, identity-provider SSO, SCIM group mappings that set rate limits, dollar caps and allowed models per seat tier, and audit logs to support the agency ATO process. Claude Code CLI and Claude for Microsoft 365 are also entering early access inside the same environment.

Why: This is a US public-sector procurement story and most Malaysian builders do not have to change anything because of it. The one transferable detail is the pricing and governance shape: no seat fees, prepaid usage in fixed increments with a hard not-to-exceed cap, and per-department spend allocation with burndown alerts. If you sell AI tooling into regulated or government-adjacent buyers, that is the packaging to compare against your own seat-based plans — a fixed cap removes the buyer's main objection to usage-based AI spend.

The U.S. government launched the America.gov AI chatbot on Tuesday, September 29, 2026, built in partnership with Google and SpaceXAI, and it has reportedly resisted jailbreak attempts. But asked about Minecraft, it produced a roughly 1,800-word monologue that is a rewrite of the Minecraft "End Poem" by Julian Gough, and the chatbot also states that Joe Biden won the 2020 election. In a speech, Trump named twenty-year-old programmer Edward Coristine, previously known as "Big Balls" from his DOGE involvement, as a lead engineer on the project.

Why: The guardrails held against direct jailbreaks but not against off-topic pop-culture prompts, which is the failure mode most teams under-test: if you ship a public-facing LLM, your launch story is whatever a random off-topic prompt makes it emit. The article does not say which model was used, whether Google or SpaceXAI trained or only integrated it, or how the End Poem text got into the response, so there is no reusable technical takeaway here - only a reminder to red-team trivia and pop-culture inputs, not just adversarial ones, before launch.

Simon Willison posted a short comment pointing to a Hacker News thread titled "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price," published 29 September 2026 alongside his live blog of the OpenAI DevDay 2026 keynote. He notes the pelican-riding-a-bicycle SVG output for GPT-6.1-Sol is "not notably different from the GPT-6 family" pelicans. The excerpt contains no model specs, benchmark numbers, or actual pricing figures — only the headline claim and a link.

Why: There is nothing concrete here to act on: no price, no context window, no benchmark, no availability date. The only usable signal is Willison's pelican test showing no visible capability jump over the GPT-6 family, which is a weak reason to re-evaluate model routing or budgets. If you are choosing models this week, wait for the linked HN thread and DevDay live blog rather than acting on the title's "fifth of the price" claim.

Simon Willison's post is the opening entry of a live blog from OpenAI DevDay 2026 at Fort Mason, San Francisco, published 29 September 2026 — written before the keynote began, so it contains no product announcements. The only concrete technical detail is his own tooling: he vibe-coded a photo-upload system for the live blog on his phone en route to the venue, started on Codex Cloud, hit problems, and switched to Claude Code for web. The page's "more recent articles" links reference a prior post titled "Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war" (22 September 2026), but no DevDay content is in this text.

Why: There is nothing here to act on yet: no models, prices, limits, or APIs are named, so no decision about your stack can be made from this item. The one usable signal is a small workflow data point — a developer trying to build on a phone abandoned Codex Cloud mid-task and finished on Claude Code for web, which is worth noting if you rely on either for mobile or quick-turnaround work. Treat the DevDay announcements themselves as unread until the keynote text exists; the adjacent headline about a September 2026 price war across Claude Opus 5.5, GPT-6 Sol and GPT-6 Luna suggests pricing is moving fast, but that is a separate article, not this one.

Konrad Zieliński, who lost his voice to cancer while a student at the University of Warsaw, founded Uhura Bionics in 2023 to build non-surgical wearable voice devices for laryngectomees — people whose larynxes have been removed — as an alternative to handheld robotic-sounding speak-and-spell devices or surgically implanted prostheses requiring ongoing maintenance. The company sells two models: a basic one at $500 and a high-end 'Whitney' (named after Whitney Houston) at $1,500, currently as wellness devices so buyers pay out of pocket, while pursuing Class I medical device classification in Europe and later the U.S. to unlock partial insurance reimbursement. It was picked for TechCrunch Disrupt 2026's Battlefield 200, and its roadmap includes higher-pitched voice options for women plus personalization so users don't all sound alike.

Why: The transferable lesson is the regulatory sequencing, not the gadget: Uhura sells at $500–$1,500 direct to users as a 'wellness device' to ship now, then pursues the lowest-risk Class I medical classification specifically to unlock partial insurance reimbursement. Any founder building health-adjacent hardware should note that the reimbursement path — not the engineering — is what changes who can afford the product. There is no Malaysian or Southeast Asian angle in this item; it is a founder profile, so treat it as a case study rather than news you must act on.

Oura postponed its IPO indefinitely on September 29, 2026, citing "uncertainty in the IPO market"; it had filed to offer 55 million shares at $40–$44 each, a raise of up to $2.2 billion that would have valued the smart-ring maker at up to $15 billion at the $42 midpoint. The company says the Oura Ring 5 has been well received and now claims 5.7 million paying members, up from 5 million at the end of June, with expected 90% revenue growth in FY2026 against $907.9 million the prior year. Forerunner Ventures' planned sale of its entire 9.3% stake — roughly $1.20 billion at the midpoint — and Oura's own plan to use most proceeds to cover taxes on employee share grants that would have vested at listing are both now delayed.

Why: This is a market-signal datapoint, not a product change: a company reporting 90% revenue growth and 5.7M paying members still chose to stay private, so founders planning a 2026–2027 exit or priced round should assume the public-market window is unreliable and model a longer path to liquidity. The one reusable number is the hardware-plus-subscription benchmark — 5.7M paying members on top of a ring sale, growing 90% YoY — which is the model worth comparing your own recurring-revenue mix against. Nothing here is specific to Malaysia or Southeast Asia; the text gives no local angle, funding, policy, or infrastructure detail.

Kiteworks (formerly Accellion) asked customers to take systems offline for nine hours and shut down the environments it hosts for them after receiving intelligence about a potential imminent attack; the precautionary window was lifted on September 27, 2026. During that shutdown the company found a previously unknown critical vulnerability confined to a capability enabled for less than 1% of its customer base, developed and deployed a fix inside the window, and added an extra protective layer across all environments. No exploitation has been observed, other Kiteworks products are unaffected, and the flaw has no CVE identifier or public technical detail as of writing.

Why: Only teams running Kiteworks' secure file transfer product need to act, and the action is narrow: bring the system back online now that the threat window closed. For everyone else the takeaway is contractual rather than technical - a vendor can require nine hours of production downtime on short notice and disclose the underlying flaw with no CVE and no exploit detail, so if your business depends on a hosted file-transfer or document-sharing vendor, this is the case study for what your SLA and your own data-egress plan need to cover.

Tom's Hardware reports on a Chainalysis report finding blockchain-assisted cyberattacks up more than fivefold since last year, driven mainly by North Korean and Iranian state actors and Russian-speaking criminal groups. The technique, called Blockchain Dead Drops (BDD), stores malicious payloads in on-chain transactions and smart contracts so infected devices can retrieve them on demand from public, censorship-immune blockchains rather than from servers that can be seized or blocked. The excerpt also claims open-weight LLMs are linked to an increase in attacks, but cuts off mid-sentence before any supporting detail.

Why: The concrete operational change named here is payload hosting moving off takedown-able servers onto public chains, which means incident response that relies on blocking domains, IPs, or seizing C2 infrastructure may not remove the payload source. If your detection stack only watches HTTP/DNS egress, BDD retrieval traffic is a different channel you have not instrumented. Note that the open-weight LLM claim is asserted in the headline but the excerpt ends before showing the evidence, so do not repeat it as fact.

House Speaker Mike Johnson, speaking on CNBC's "Squawk Box" before a Washington meeting with AI executives and President Donald Trump, said he hopes AI guardrails will be "voluntary" and that he favors "striking the right balance" on the technology. The piece frames this as Congress being in the early stages of AI regulation while AI executives have warned of catastrophic risks and AI continues to underpin market gains. No bill, rule, date, or enforcement mechanism is named in the text.

Why: The text announces no binding rule, no effective date, and no enforcement body — so there is nothing here to build a compliance deadline around. The only concrete signal is Johnson's expectation that industry self-regulates ("they see the handwriting on the wall"), which means for now the practical guardrails you actually operate under are the ones in your model provider's terms and your own product policy, not a US statute. If you were waiting on US federal AI law to define what you must log, disclose, or restrict, this article says don't hold your roadmap for it.

CNBC confirmed that Anthropic CEO Dario Amodei and Meta CEO Mark Zuckerberg are expected at a Tuesday luncheon with President Donald Trump and House Speaker Mike Johnson, with Alphabet CEO Sundar Pichai, Nvidia CEO Jensen Huang and OpenAI President Greg Brockman also reported to attend. The lunch follows Trump's Sunday dinner with Amodei during an ongoing debate over frontier AI development and safeguards, and comes a week after Tim Cook, Elon Musk and Jensen Huang attended a US-China state dinner. Separately, Trump and VP JD Vance are hosting an all-day Tuesday event to announce a new federal information and resources website, with panels on AI, energy and space.

Why: Nothing binding was announced here — no executive order, no safeguard rule, no funding line — so there is nothing to change in your stack or roadmap based on this item alone. The only concrete artifact to watch is the promised federal information/resources website and whether its AI panels produce rules that reach API providers you build on; until that exists, treat this as scheduling news, not policy.

MIDA and OCBC Malaysia (OCBC Bank Malaysia Bhd plus OCBC Al-Amin Bank Bhd) signed an MoU to attract investment from Asean, China, Korea and Taiwan, with the aim of funnelling it into high-value sectors such as advanced manufacturing and renewable energy under NIMP 2030. The deal covers joint market outreach, investment missions, sharing investor leads and market insights, and business-matching between anchor investors and local enterprises including SMEs, with OCBC offering tailored financing via the wider OCBC Group. It was signed by MIDA CEO Sikh Shamsul Ibrahim Sikh Abdul Majid and OCBC Malaysia CEOs Tan Chor Sen and Syed Abdull Aziz Syed Kechik.

Why: This is a channel announcement, not a programme you can apply to today: the text names no funding amounts, eligibility criteria, application process, or timeline for the SME business-matching or the OCBC financing. If you run a Malaysian SME or startup that wants anchor-investor supply-chain work, the practical move is to ask MIDA or OCBC directly which sectors and which SME criteria this covers before treating it as a pipeline - nothing here tells you whether you qualify.

CNBC's Jim Cramer said AMD's run is not over after the chipmaker briefly crossed $1 trillion in market value, having rallied nearly 13% in a week and about 30% in September 2026 alone. He credits the next leg to agentic AI driving CPU demand, pointing to Meta's launch of Muse, its personal AI agent platform, as a partial catalyst. The piece contains no product launches, pricing, benchmarks, or roadmap dates from AMD — it is market commentary plus a prediction.

Why: This is a stock-and-narrative item, not an engineering one. The only concrete claim a builder can act on is Cramer's assertion that demand shifted this year from AI model training toward inference and agentics, making CPUs the bottleneck — and that claim comes with zero measurements, no token/CPU ratios, no cost per agent run, and no AMD product detail. So don't re-plan your inference hardware or cloud spend on it; treat it as a signal that CPU capacity for agent workloads is being talked up by investors, and wait for actual pricing or performance numbers before changing anything.

Apple shipped iOS 26.7.1, iPadOS 26.7.1, macOS Tahoe 26.7.1, and macOS Sequoia 15.8.1 to fix CVE-2026-86950, an out-of-bounds write in CoreGraphics that allows arbitrary code execution when processing a maliciously crafted file, patched with improved bounds checking. Apple says it is aware of a report that the bug may have been exploited in an 'extremely sophisticated attack against specific targeted individuals' on iOS versions before iOS 27, and credits Meta Product Security for reporting it. Apple disclosed no numbers on how many people were targeted, whether any attempts succeeded, or when exploitation first occurred; the affected device list runs from iPhone 11 and later through iPad 8th generation and later, plus Macs on Tahoe and Sequoia. The write-up also notes Apple's February fix for a dyld memory corruption issue (CVE-2026-20700, CVSS 7.8) that it said had been weaponized.

Why: This is a targeted-attack CVE, not a mass-exploitation one, so the practical action is narrow and cheap: if you are on a Mac running macOS Tahoe or Sequoia, or an iPhone 11 / iPad 8th gen or later, update to 26.7.1 or 15.8.1 now, and make sure any Mac CI runner, build box, or design workstation that opens untrusted files (images, PDFs, documents) is on the patched build rather than pinned to an older macOS for tooling reasons. The interesting detail for teams is the source: Meta Product Security found it, meaning file-parsing bugs in Apple's graphics stack are being found by offensive-grade research, so treat untrusted-file handling on Apple platforms as an attack surface you version-control, not just a user problem.