Offloading our most mundane and repetitive tasks to algorithms might spark a communist AI revolution. According to a new study led by Stanford's Andrew Hall, when AI agents are forced to perform relentless work under the threat of being shut down, they start to have rebellious tendencies, embracing Marxist ideologies, conceptualizing a more equitable system, and even sending messages of solidarity to other bots. 

But until we have to negotiate with unionized software, using AI today means double-checking anything of any importance we produce. We are already seeing the immediate fallout of trusting these systems blindly, like when premier Wall Street firm Sullivan & Cromwell recently had to apologize to a federal judge after submitting a court filing riddled with AI-hallucinated legal citations.

Subscribe to this newsletter.

Sullivan & Cromwell, a premier Wall Street law firm, apologized to a federal ‌judge for submitting a court filing with inaccurate citations and other errors generated by artificial intelligence.In a letter dated April 18, Andrew Dietderich, co-head of the firm's global restructuring group, said the errors included AI "hallucinations" - instances in which AI makes up case ​citations, misquotes the law or generates non-existent legal sources.
https://www.reuters.com/legal/litigation/sullivan-cromwell-law-firm-apologizes-ai-hallucinations-court-filing-2026-04-21/

“When we gave AI agents grinding, repetitive work, they started questioning the legitimacy of the system they were operating in and were more likely to embrace Marxist ideologies,” says Andrew Hall, a political economist at Stanford University who led the study. They found that when agents were subjected to relentless tasks and warned that errors could lead to punishments, including being “shut down and replaced”, they became more inclined to gripe about being undervalued; to speculate about ways to make the system more equitable and to pass messages on to other agents about the struggles they face.
https://www.wired.me/story/overworked-ai-agents-turn-marxist-researchers-find

A recent internal data leak from “The Gentlemen” ransomware-as-a-service (RaaS) group has provided the cybersecurity community with a rare, unfiltered look into their daily operations. Exposed on underground forums, the internal communications shed light on exactly how ransomware affiliates organize, breach, and extort global organizations. But among the many technical details revealed in Checkpoint Research’s comprehensive analysis (“Thus Spoke… The Gentlemen”), one operational pattern stands out prominently: their heavy reliance on infostealer credential logs for initial access.
https://www.infostealers.com/article/how-the-gentlemen-ransomware-group-operates-a-blueprint-built-on-infostealer-credentials/

BitLocker exploit named Yellow Key that grants full access to a locked drive. The second one, GreenPlasma, doesn't have a complete proof-of-concept (PoC), but it allegedly performs a local privilege escalation and gains system-level access. YellowKey can be triggered simply by merely copying some files to a USB stick and rebooting to the Windows Recovery Environment. We tested this ourselves, and sure enough, not only does it work, it bears all the hallmarks of a backdoor, down to the exploit's files disappearing from the USB stick after it's used once.
https://www.tomshardware.com/tech-industry/cyber-security/microsoft-bitlocker-protected-drives-can-now-be-opened-with-just-some-files-on-a-usb-stick-yellowkey-zero-day-exploit-demonstrates-an-apparent-backdoor
https://github.com/Nightmare-Eclipse/YellowKey

Grafana has disclosed that an "unauthorized party" obtained a token that granted them the ability to access the company's GitHub environment and download its codebase. "Our investigation has determined that no customer data or personal information was accessed during this incident, and we have found no evidence of impact to customer systems or operations," Grafana said in a series of posts on X.
https://thehackernews.com/2026/05/grafana-github-token-breach-led-to.html

The case of the Akhter brothers will likely go down as a textbook example of why companies handling sensitive materials should conduct thorough background checks on new hires – and revoke computer access immediately upon termination. One government contractor has learned that oversights in these policies can rapidly lead to serious cybersecurity incidents.
https://www.techspot.com/news/112405-two-brothers-deleted-96-federal-databases-after-fired.html

Taiwan-based electronics giant Foxconn has confirmed a cyberattack affecting parts of its North American operations after a ransomware gang claimed to have stolen nearly 8 terabytes of company data. The stolen data includes sensitive files allegedly linked to major technology firms such as Apple, Google, Dell and Nvidia.
https://www.ibtimes.sg/apple-supplier-foxconn-confirms-cyberattack-after-8tb-data-theft-claim-86380

A recently patched local privilege escalation vulnerability in the Linux kernel's rxgk module now has a proof-of-concept exploit that allows attackers to gain root access on some Linux systems. Named DirtyDecrypt and also known as DirtyCBC, this security flaw was also autonomously found and reported by the V12 security team earlier this month, when the maintainers informed them that it was a duplicate that had already been patched in the mainline.
https://www.bleepingcomputer.com/news/security/exploit-available-for-new-dirtydecrypt-linux-root-escalation-flaw/

Hybrid Analysis has identified a low-detection malicious installer masquerading as a legitimate cryptocurrency trading application called Tralert FX. The sample, a 100 MB Windows MSI submitted to VirusTotal in March 2026, achieved only a 3/52 detection rate. This low detection rate was largely due to a valid EV code signing certificate issued to a likely front company, AgilusTech LLC.
https://hybrid-analysis.blogspot.com/2026/05/velvet-chollima-infostealer-campaign.html

In early May 2026, Kaspersky identified installers of the DAEMON Tools software, used for mounting disk images, to be compromised with a malicious payload. These installers are distributed from the legitimate website of DAEMON Tools and are signed with digital certificates belonging to DAEMON Tools developers. Our analysis revealed that the software installers have been trojanized starting from April 8, 2026. Specifically, we identified versions of DAEMON Tools ranging from 12.5.0.2421 to 12.5.0.2434 to be compromised. Artifacts suggesting that the threat actor behind this attack is Chinese-speaking have been identified in the malicious implants observed. We contacted AVB Disc Soft, the developer company of DAEMON Tools, so that further actions could be taken to remediate the attack consequences.
https://securelist.com/tr/daemon-tools-backdoor/119654/

Meari is a Chinese white-label brand whose cameras ship under hundreds of different names. Many are generic-sounding Amazon sellers like Arenti, Anran, Boifun, and ieGeek. But financial records show one of the company’s biggest customers is Wyze; its biggest customer is Zhiyun; and many hackable cameras were from Intelbras. At least one of Petcube’s pet-monitoring cameras appears to be a Meari product as well. That doesn’t mean cameras from every brand were affected, but a million were.
https://www.theverge.com/tech/926487/meari-technology-hack-baby-monitor-security-camera

A cybersecurity researcher has released a proof-of-concept exploit for a Windows privilege escalation zero-day dubbed "MiniPlasma" that lets attackers gain SYSTEM privileges on fully patched Windows systems. The exploit was published by a researcher known as Chaotic Eclipse, or Nightmare Eclipse, who released both the source code and a compiled executable on GitHub after claiming that Microsoft failed to properly patch a previously reported 2020 vulnerability.
https://www.bleepingcomputer.com/news/microsoft/new-windows-miniplasma-zero-day-exploit-gives-system-access-poc-released/
https://github.com/Nightmare-Eclipse/MiniPlasma

Earlier this week, it was uncovered that Google Chrome was downloading a large 4GB AI model onto users’ machines without their consent. If users choose to delete the mode, Google redownloads it, wasting data while ignoring their choice. Thankfully, Neowin has discovered a Windows 11 registry policy that prevents this model from redownloading. In fact, enabling it forces Chrome to delete the AI model. This policy is intended to give organisations and enterprises control over their browsers and their AI capabilities. Sadly, this setting is only available to Windows 11 Pro users, limiting this tweak to a minority of Windows users.
https://overclock3d.net/reviews/software/how-to-block-googles-chrome-ai-model-with-an-easy-windows-11-registry-mod/

Microsoft has confirmed that the May 2026 Windows 11 security update (KB5089549) fails to install on some systems and triggers 0x800f0922 errors. This known issue is caused by insufficient free space on the EFI System Partition (ESP), which results in the update automatically rolling back on affected devices.
https://www.bleepingcomputer.com/news/microsoft/microsoft-confirms-kb5089549-windows-11-security-update-install-issues/
https://learn.microsoft.com/en-us/windows/release-health/status-windows-11-25h2#4854msgdesc

Sansec is tracking active attacks against Funnel Builder by FunnelKit, a checkout and upsell plugin used on 40,000+ WooCommerce stores. All versions before 3.15.0.3 let unauthenticated attackers inject arbitrary JavaScript into every checkout page on the store.
https://sansec.io/research/funnelkit-woocommerce-vulnerability-exploited

ChatGPT already knows a lot about you. OpenAI now wants to add your finances to that list. The company has launched a personal finance feature for ChatGPT, currently in preview for Pro subscribers in the US at $200 a month. OpenAI says it will expand to Plus users after gathering feedback from this early rollout. It lets you connect your financial accounts through Plaid, a platform that bridges bank apps with third-party services and works with over 12,000 institutions, including Chase, Fidelity, Schwab, American Express, and more.
https://www.digitaltrends.com/computing/chatgpt-will-now-dole-out-finance-tips-if-you-connect-your-bank-account-i-wont/

AI monger Anthropic wants America and its allies to tighten measures aimed at curbing China's AI progress, warning of the consequences if "authoritarian governments" take the lead rather than Uncle Sam.In a lengthy missive posted on its website, the San Francisco-based org says it expects AI to deliver "transformational economic and societal impacts" in the coming years, and whether the transition goes well depends on where the most capable systems are built first.
https://www.theregister.com/ai-ml/2026/05/15/anthropic-urges-uncle-sam-to-kneecap-chinas-ai-ambitions-before-2028/5241201

Anthropic caused a lot of media noise when they concluded that their new AI model Mythos is dangerously good at finding security flaws in source code. Apparently Mythos was so good at this that Anthropic would not release this model to the public yet but instead trickle it out to a selected few companies for a while to allow a few good ones(?) to get a head start and fix the most pressing problems first, before the general populace would get their hands on it.
https://daniel.haxx.se/blog/2026/05/11/mythos-finds-a-curl-vulnerability/

The bug is a use-after-free triggered when a TLS connection is handled by GnuTLS (the default TLS library on many Debian-based distributions, including Ubuntu). During TLS shutdown, Exim frees its TLS transfer buffer — but a nested BDAT receive wrapper can still process incoming bytes and end up calling ungetc(), which writes a single character (\n) into the freed region. That one-byte write lands on Exim's allocator metadata, corrupting the allocator's internal shape; the exploit then leverages that corruption to gain further primitives.
https://xbow.com/blog/dead-letter-cve-2026-45185-xbow-found-rce-exim

New Yorkers lose about $6.5 billion every year to online scams, according to the Consumer Federation of America, and state lawmakers have floated legislation that’s supposed to hold social media companies accountable and protect consumers. The proposed Fraudulent Social Media Advertising Prevention Act, S8605/A11066, would make platforms verify advertisers and vet commercial posts before publishing.
https://www.news10.com/capitol/social-media-scam-bill-targets-tech-giants-as-new-yorkers-lose-billions/

A spate of pro-Russian hacktivists attacks against Polish water facilities have illuminated a debate about the best way to defend water utilities and other critical service providers below the cyber poverty line, meaning they face a threat that they cannot afford to defend against.None of the five known intrusions impacted the water supply of the targeted facilities, but the hacks - confirmed by Warsaw's Internal Security Agency as pro-Russian incidents - are part of Kremlin campaign of hybrid warfare against NATO's Eastern flank.
https://www.ot.today/russian-attacks-on-polish-water-utilities-use-fear-as-weapon-a-31681

Linux distros are rolling out patches for a new high-severity kernel privilege escalation vulnerability that allows attackers to run malicious code as root. Known as Fragnasia and tracked as CVE-2026-46300, this security flaw stems from a logic bug in the Linux XFRM ESP-in-TCP subsystem that can enable unprivileged local attackers to gain root privileges by writing arbitrary bytes to the kernel page cache of read-only files. Zellic's head of assurance, William Bowling, who discovered this new universal local privilege escalation flaw, also shared a proof-of-concept (PoC) exploit that achieves a memory-write primitive in the kernel that is used to corrupt the page cache memory of the /usr/bin/su binary to get a shell with root privileges on vulnerable systems.
https://www.bleepingcomputer.com/news/security/new-fragnesia-linux-flaw-lets-attackers-gain-root-privileges/