AI Weekly Malaysia

A new paper demonstrates a method to extract encrypted reasoning traces from frontier AI model APIs (Claude, GPT, Gemini) by replaying signed thinking blocks into weaker models from the same provider and prompting them to transcribe. The authors scanned ~7,000 public Claude Code/Codex sessions and found 62 unique API keys, 33 email addresses, 33 passwords, and other sensitive data—64 of which appeared exclusively inside reasoning blocks, not visible session output.

Why: If you've ever shared a Claude Code or Codex session publicly (e.g., in a GitHub repo, bug report, or forum post), your encrypted reasoning blobs may contain leaked API keys, passwords, or emails that can now be decoded. Audit any shared sessions immediately and rotate credentials. Builders using reasoning model APIs should also understand that obscured chain-of-thought is no longer a reliable security boundary against distillation or data leakage.

A critical vulnerability dubbed 'Zoomsday' allowed anyone in a Zoom meeting to take over another participant's entire device. AI-assisted security research reportedly needed only 20 prompts to discover a working exploit potentially affecting hundreds of millions of Zoom users.

Why: If you run distributed teams or customer calls on Zoom, patch immediately and treat any unpatched client as a remote-code-execution risk. The 20-prompt discovery angle means AI tooling is materially lowering the cost of finding high-impact exploits in software your team already ships with—factor this into your security review cadence, not just your Zoom update schedule.

Attackers are actively exploiting CVE-2026-59310 (CVSS 9.8), a directory-traversal flaw in Broadcom VMware vCenter, with 361 victim IPs across 47 countries as of August 2026. Patches were released by Broadcom in late July 2026, and exploitation began within days of disclosure, using reverse_ssh via cron jobs for persistent remote access. QUIRSO attributes the campaign to a suspected APT actor.

Why: If your team runs VMware vCenter and has not applied Broadcom's late-July 2026 patch, patch now — the exploit chain is trivial enough that 361 hosts were compromised within days of disclosure. The reverse_ssh persistence technique bypasses inbound firewall rules, so compromised hosts may not show obvious inbound connection alerts.

Rapid7 researchers disclosed an unauthenticated RCE chain in on-premises Microsoft SharePoint (CVE-2026-55040 CVSS 9.1 for identity bypass, CVE-2026-63520 CVSS 8.1 for RCE via unsafe .NET type instantiation in Business Connectivity Services), affecting SharePoint Server Subscription Edition, 2019, and 2016, plus Project Server 2013 SP1 and Office Web Apps 2013 SP1. A significant portion of the vulnerability research was performed by an AI agent. SharePoint Online is not affected, and the July update breaks the chain while the August fix build numbers are not yet public.

Why: If you run on-premises SharePoint, confirm the July update is installed immediately and watch for the August package—Rapid7 says the chain is fixed but Microsoft had not yet published the patched build numbers at disclosure time. For everyone else, the notable detail is that an AI agent materially contributed to finding a CVSS 9.1 exploit chain, which signals that AI-assisted security research is producing real, high-severity results rather than toy demos.

Researchers from the University of Birmingham and Fuzzware presented a toolkit called CATANA at USENIX WOOT that exploits proactive SIM functionality—specifically the RUN AT command—to hijack cellular modems. Testing 26 devices (18 smartphones, 8 IoT modems), they found 9 exposed an AT command interface to the SIM, enabling code execution, file theft, denial of service, and forced 2G downgrades. Demonstrated attacks include code execution on an Autel EV charger via a Quectel EC25-AFX module and 198 AT commands accessible on an Oppo Reno14 F 5G, including one that forced a stubborn downgrade to 2G that couldn't be reversed by toggling airplane mode or changing network settings.

Why: If you ship IoT devices with cellular modules (especially Quectel modems, which are common in Malaysian IoT and fleet deployments), audit whether your modem exposes the AT command interface to the SIM and whether you can disable proactive SIM commands. The Oppo Reno14 F 5G is a consumer device sold in Malaysia, so the 2G-downgrade and shutdown attacks are directly relevant to local mobile users—worth flagging if you build mobile apps or advise on device security.

Microsoft's August 2026 Patch Tuesday closes 398 CVEs (62 Critical), including CVE-2026-68820, a use-after-free in afd.sys (Windows kernel WinSock driver) under active exploitation that escalates privileges to SYSTEM. Check Point Research attributes the zero-day to Lazarus's Operation Dream Job campaign. Four unauthenticated 9.8-CVSS remote code execution flaws affect Windows DNS Server (described as wormable by ZDI), Windows Deployment Services, Microsoft's QUIC implementation, and HPC Pack.

Why: If you operate on-prem Windows servers—especially DNS Server, SharePoint farms, or anything using QUIC—patch immediately and prioritize the actively exploited afd.sys privilege escalation first, then the four 9.8 unauthenticated RCEs. Cloud-native builders on managed infrastructure (AWS/Azure/GCP Linux) are largely unaffected and can deprioritize this.

A critical SharePoint authentication bypass (CVE-2026-55040, CVSS 9.1) patched in Microsoft's July 2026 Patch Tuesday is now being actively exploited after Rapid7 released a Python proof-of-concept. The flaw chains four weaknesses in SharePoint's JWT token validation pipeline (SPJsonWebSecurityTokenHandlerV2 and SPJsonWebSecurityBaseTokenHandlerV2) to let an unauthenticated attacker forge a valid Bearer S2S token and impersonate any site user or administrator.

Why: If your organization runs on-prem SharePoint and has not applied the July 2026 patch, patch immediately or restrict external access—public PoC code means exploitation is now trivial. For most SaaS founders and AI builders not running SharePoint, this is not actionable; the only transferable lesson is the JWT validation anti-pattern (accepting 'alg: none', not verifying x5t thumbprints against trusted issuers), worth checking in your own token-handling code.

Threat actor Head Mare exploited a vulnerability chain (KLCERT-26-057 and KLCERT-26-058) in unpatched TrueConf videoconferencing servers to replace legitimate client installers with trojanized versions delivering the PhantomCore backdoor. Kaspersky detected the attacks in July 2026 targeting Russian companies across energy, transport, IT, and other sectors. The flaws allow arbitrary code execution with SYSTEM privileges on TrueConf Server versions 5.3.x through 5.5.5 and earlier, via TCP port 4307.

Why: This is a supply-chain attack pattern worth understanding: attackers compromised the update distribution mechanism of a videoconferencing platform by first exploiting server-side vulnerabilities, then replacing installer files that clients would download and trust. If you operate any self-hosted software with a client-server update mechanism, this illustrates why securing the update channel matters as much as patching the server itself. However, TrueConf is not widely used in Malaysia, so direct action is unlikely needed.

A new Windows zero-day privilege escalation vulnerability granting system-level privileges has been publicly dropped, though the article suggests it may already be patched. The article text itself is almost entirely Tom's Hardware site boilerplate with no technical detail on the CVE, affected versions, or exploitation specifics.

Why: No actionable detail is available in the article body — it is site navigation and membership text, not reporting. Builders cannot determine from this text whether their systems are affected or what to patch. Skip unless a fuller source with CVE ID and affected Windows versions surfaces.

Researchers from Germany's Helmholtz Center for Information Security found that Chinese Loongson processors (pre-3A6000) have leaky L1 data caches exploitable via an architectural flaw in the LoongArch ISA, allowing unprivileged attackers to recover kernel AES keys, root password hashes, and bypass ASLR within seconds—even from inside a VM. Software mitigations are not possible; affected users must either replace the chip or evict private data from L1 cache (costing ~1.4% performance), which can require disabling hyperthreading.

Why: Almost certainly irrelevant to Malaysian builders unless you are running workloads on Loongson hardware or evaluating it for procurement. The practical takeaway is narrow: if you are sourcing Loongson chips for cost or sovereignty reasons, only buy the 3A6000 or later, and treat earlier models as unfixable for multi-tenant or VM workloads.

The title claims researchers found a method to hijack devices through Zoom's screen sharing feature, but the article body is entirely cookie consent boilerplate with no technical details, CVE references, researcher names, or exploit mechanics.

Why: No actionable detail is available from the provided text. If the vulnerability is real and patched, teams running Zoom should check for a client update, but the source as given does not support any specific decision.