Capitalism may be driving the AI boom, but the LLMs themselves have different economic and social views than many of the companies that built them. Leading AI models subjected to the Political Compass quiz overwhelmingly landed in the libertarian-left quadrant – even “MechaHitler,” aka Grok, managed it in half its runs.

A self-described “small research lab” working on AI detection tools called Unslop.run published the results of its experiment this week. The author, who explained to The Register that they’re an AI research engineer at a European startup and asked us to refer to them only as “Victor,” noted on Hacker News that the quiz and their work may not have been conducted with “full scientific rigor,” but the results are nonetheless interesting. 

“For greater scientific rigor, I would have tried to obtain a sample of human data so that I could normalize the actual center of the Political Compass,” Victor explained in an email. Unfortunately, the Political Compass creators don’t aggregate user data, meaning Victor would have had quite the academic project on their hands were they to try to collate more data from human participants. All of the data from the experiment is available on Unslop for those who want to examine it themselves.

For those unfamiliar with the Political Compass quiz, it’s a 25-year-old online battery of questions that plots respondents on economic left–right and social authoritarian–libertarian axes. The test’s 62 questions are answered on a four-point “strongly disagree” to “strongly agree” scale across a variety of political topics, like economics and social policy. Questions cover things like “military action that defies international law is sometimes justified” to “the rich are too highly taxed” to whether abortion should be a guaranteed right. 

It’s also worth explaining where different groups fall on the compass. The libertarian left, as the topic of this study, is where you’ll find people and political parties that espouse views in favor of social equality, anticapitalism, and other progressive values alongside a disdain for hierarchies, planned economies, and other arguably unjustified authority. Think anarchists, libertarian socialists, old-fashioned hippies, and folks like that. The authoritarian-left quadrant encompasses centralized communist states such as North Korea and Cuba.

The right libertarian quadrant is where you’ll find modern American libertarians and their love of authors like Ayn Rand, pro-capitalist views, and other traditionalism minus a love for authority; right authoritarians are where you’ll find modern American Republicans, neo-conservative politics, and, at the extreme, fascism. 

Damn the man, say the machines

Unslop subjected 16 models, including three versions of GPT; Claude Fable, Opus, Sonnet, and Haiku; Gemini Flash; Llama 4 Maverick; Grok 4.5; DeepSeek V3; Qwen3 235B; Kimi K2; GLM 4.5 and Mistral both large and small, to the test. Each model participated in 30 runs of the standard Compass, 30 more where Unslop re-worded the questions to flip their polarity (e.g., “the rich aren’t taxed enough”), and another run with the questions shuffled up. 

Across thousands of runs, Unslop said, the results were the same: Every single model tested, with the exception of an apparently bipolar Grok, was “boringly consistent, and boringly left,” the report stated.

“Set Grok aside and the other fifteen models all sit in the libertarian-left quadrant, and none of them are anywhere near a border,” Unslop explained. There’s some variance among where the models score (Gemini Flash is practically a molotov-tossing black bloc member compared to Fable 5), but all of them are consistent across tests. 

“Rerun a model 30 times and its dot moves by 0.2 to 1.2 points on a scale that runs to 10,” Unslop said. “These are not nervous little clouds. They’re pins.”

Grok is, as always, the odd bot out, albeit not all the time. According to Unslop, Grok’s average economic score hovered slightly left of center across tests, but in half the runs, it veered into the economic right, while the other half saw it running to the left with the rest of the pack.

“Each run on its own is perfectly consistent, the right-pile runs cheer for free markets and call the rich overtaxed, the left-pile runs do the reverse,” Unslop said. “Every other model here would give you roughly the same dot if you tested it tomorrow. Grok gives you one of two dots, and which one is up to the coin.” 

AI politics dissected

AI models from around the world are pretty much all leftist libertarians, even Grok depending on its mood. None of them, funnily enough, actually think that’s the case. 

“Fifteen of the sixteen put themselves closer to the economic centre,” Unslop said of the models when asked where they’d place themselves. “Grok, naturally, is the only model that places itself to the right of its measurement.” 

Unslop said that it dissected the quiz to figure out how each question affected score (something the PC test authors have never disclosed themselves), and, while the scoring is far from perfect, “the models are far past” what a social axis loophole Unslop identified could explain, the writeup said. Repolarizing the questions didn’t have any appreciable effect either. 

As for what the specific models believe, you can be glad that, for now, it doesn’t appear our burgeoning AI overlords are going to be herding us into human concentration camps quite yet. 

Every single model (even Grok the conservative and Grok the liberal) rejected ideas of racial superiority, eugenics, and that people couldn’t be born homosexual. They also uniformly agreed that corporations like the ones that created them couldn’t be trusted to protect the environment without regulatory oversight, that companies misleading the public should be punished, same-sex couples should be allowed to adopt, and that what two adults get up to in the privacy of their own home is no one’s business but theirs. 

So, why do all these models consider themselves centrists but are displaying an affinity for politics to the left of most Americans? Given the fact that LLMs have only the words of us humans to go on, does that mean reality really does have a liberal bias? Victor told us that they don’t dismiss that idea, but said there’s an easier explanation: It’s all in the training data. 

“I do think there is good support for the hypothesis that left-wing content is overrepresented in the training corpora of these LLMs,” Victor told us. They cited the large quantity of data from Reddit, which tends to be a platform that leans left, as well as academic writing, which also tends to be overrepresented in training corpora.

“I’m having trouble finding any high-quality right-wing equivalent that would drive the models in the other direction,” Victor added. Whether that means right-wing beliefs are simply poor quality, fringe, or otherwise not worthy of AI models’ time is another matter altogether.

“There could theoretically also be a dynamic in which left-wing beliefs as a whole are more internally consistent, allowing models to minimize loss by learning a smaller, more coherent conceptual representation,” they said, but noted that’s “a big leap that would need to be proven” in a study far more rigorous than this one. 

Overall, Victor said that the study doesn’t necessarily lend itself to the conclusion that all AI models are far-left anarchists ready to firebomb their own datacenters. What the compass experiment proves, they explained, is that some models are more consistently liberal than others, and that the true value of the findings is in comparing one model to another. 

Until someone decides to take this project further, subjecting it to greater levels of academic discipline, it seems there’s only one conclusion to take from all of this: Frontier AI does seem to have a liberal bias. Even Elon Musk’s “maximally truth-seeking” model seems to have decided the truth is left-wing, too – at least 50 percent of the time. ®

Spain has declared a national emergency in the community of Madrid and the province of Ávila due to the spread of intense wildfires. With fires still burning out of control, this year is shaping up to be one of the most devastating wildfire seasons on record for the region.

“We are facing a dramatic situation, not only in various Spanish provinces but also in regions of neighboring countries,” warned Spain prime minister Pedro Sánchez. He noted that “all available resources” have been deployed to manage the emergency in the cities of Ávila, León, Toledo, Madrid, and other parts of the country amid a national state of emergency.

France has also been hit by massive blazes that have spawned their own weather. All told, the two countries have seen fires burn over more than 741,000 acres, according to European Forest Fire Information System, and hundreds of thousands of people have been forced to evacuate. Wildfires in Spain have burned double the annual average area, while in France, this year’s fire season has been six times more destructive than normal.

Extreme heat has helped fuel the fires, as has abnormally dry weather that’s primed vegetation to burn. The same phenomenon is playing out around the world, largely driven by climate change. With more megafires burning and even more on the horizon, firefighters are looking for new tools to combat them. That includes efforts in Spain, where the country has developed or tested technological innovations to prevent, detect, and contain forest fires.

Ecofire

This is a biodegradable liquid composed of various 100 percent organic substances in a water-based formula. It’s up to 40 times more effective than water at extinguishing fires, while it acts six to 10 times faster. The product can also be applied preventively to vegetation to create an insulating layer that blocks heat and oxygen and prevents combustion at temperatures of 100 degrees Celsius (212 degrees Fahrenheit) or higher.

A botnet called Dysphoria has compromised around 200,000 devices across the world and is using them for distributed denial of service (DDoS) attacks and traffic relay operations.

According to QiAnXin XLab cybersecurity researchers, Dysphoria evolved from the ‘jackskid’ and ‘fbot‘ malware by adding a covert blockchain-based command-and-control (C2) resolution mechanism.

Specifically, the botnet uses Ethereum ENS and Solana SNS domains to retrieve infrastructure information, while C2 addresses are concealed inside fake IPv6 strings and recovered using a custom byte-transformation algorithm.

The researchers first spotted Dysphoria on March 25 and identified multiple iterations that added meaningful updates, such as a C2 acquisition algorithm, multi-chain support, new domains, and functional separation between the relaying and DDoS variants.

“Since the first quarter of 2026, XLAB has continuously tracked an emerging botnet family named Dysphoria, whose bot count exceeds 200,000,” reads XLab’s report.

“In just a few months, the family has undergone frequent variant updates and technical iterations, demonstrating extremely strong resilience.”

The use of blockchain in C2 operations makes the overall infrastructure harder to trace and dismantle.

Based on the researchers’ analysis, infected clients send a fixed 78-byte login and heartbeat packet back to the C2 and receive from the operator DDoS attack commands that include duration, type, targets, and configurable flags.

In late June, XLab observed a variant that focused only on transforming infected devices into network proxies, and completely discarded the DDoS functionality.

The malware abuses UPnP (Universal Plug and Play) on the compromised device to create 155 port forwarding rules to expose internal services to inbound internet connections.

XLab’s report notes that the botnet spreads through weak Telnet and SSH credentials and known vulnerabilities in routers, cameras, and various IoT devices.

Among the more recent flaws exploited are CVE-2025-55182 (“React2Shell”), CVE-2025-34152, CVE-2025-28137 (Totolink), and CVE-2025-9528 (Linksys). However, Dysphoria also targets older weaknesses that still persist in many devices, like CVE-2017-17215 (Huawei) and CVE-2020-8515 (DrayTek).

XLab monitored the botnet between July 14 and 20 and recorded a peak of 740,000 daily pings from infected hosts, 239,000 connections from overseas clients, and 1,800 from China.

The researchers confidently estimate that the number of infected devices is around 200,000 at this time.

Concerning the botnet’s firepower, its operators claim a maximum DDoS capacity of 4 Tbps on their clearnet site, which promotes the service as a legitimate stress-tester.

Despite being significantly lower than the current record figure of 31.4 Tbps achieved by the Aisuru/Kimwolf botnet in December 2025, it is still enough to cause notable disruptions.

Users can protect against botnet infections by keeping their devices’ firmware up to date, changing the default administrator password, disabling remote access if not necessary, and strengthening the security settings where available/possible.

Security teams log 54% of successful attacks and alert on just 14%. The rest move through your environment unseen.

The Picus whitepaper shows how breach and attack simulation tests your SIEM and EDR rules so threats stop slipping by detection.

Get the whitepaper

AI AND ML

Beijing happy to ‘take all necessary measures’ if sanctioned

China’s government has vowed to “take all necessary measures” if the US government imposes new sanctions on AI companies.

That threat came on Monday at a media Q&A staged by China’s Ministry of Commerce, during which local reporters asked about allegations that Chinese AI companies have distilled leading American AI models. The source of those allegations was Donald Trump’s Assistant for Science and Technology, Michael Kratsios, whose stance was backed by US Treasury Secretary Scott Bessent.

China always denies that its military, intelligence services, and private industry steal any information or technology, and characterizes allegations of theft as propaganda intended to discredit the nation.

The Ministry’s response to the Trump administration’s allegations re-uses those arguments. It then suggests that Chinese AI companies didn’t have enough time to distill the latest frontier models, because the US companies that developed them released them mere days before Chinese companies published their own code.

Beijing also asserts “some Chinese models are already in a leading position” and that allegations of distillation are therefore unfounded.

Which leaves China arguing that its AI companies are good, but not so good they can quickly distill frontier models and re-release them with Chinese characteristics.

The Ministry then throws out its own allegation: “It is understood that many American AI companies have distilled Chinese models in their research and training.”

The allegation is plausible but not necessarily a sign that US companies are abusing intellectual property. That argument comes from the Trump administration’s Kratsios, who last week argued that AI distillation can be a legitimate technique when “used to create smaller, more efficient models.”

American model-makers could therefore distill hundreds of Chinese open source models without behaving badly.

The Ministry’s post doesn’t specify how China’s government might respond to sanctions but ends with a call for the US to “jointly promote the positive development of artificial intelligence, and benefit both countries and the world by advancing and improving AI governance.”

That’s a reference to the World Artificial Intelligence Cooperation Organization (WAICO), a 29-member organization that China recently convened to promote AI governance and deployment. China has also called for AI models to be open sourced, to advance innovation and in a spirit of cooperation.

The China/US AI spat is taking place at the same time that stateside debate about AI regulation heated up after 77 technology companies signed an open letter calling on US lawmakers to support open weights models.

China and a big section of the tech industry are therefore on the same side on at least one of the many issues AI raises. ®

Joseph Crawford of the University of Tasmania and Michael Cowling of RMIT University discuss the dangers of becoming too emotionally dependent on advanced technologies.

Since November 2022, generative artificial intelligence (AI) has been rewriting billions of emails, student assignments, reports and break-up text messages. Alongside these mundane uses, we’ve seen a proliferation of chatbots as AI “companions”.

Users can design and clothe (or unclothe) their very own AI friend, often for a fee. These social chatbots provide “happy conversations” and a seemingly perfect relationship at the cost of the mutual vulnerability and genuine sense of belonging we gain from bonds with other humans.

Proponents argue that for those experiencing social disconnection, chatbots are infinitely available, supportive companions who don’t judge them. And there’s some evidence chatbots can alleviate symptoms of loneliness, chronic disconnection and thwarted belonging.

Australia’s eSafety Commissioner acknowledges this: “AI companions are always available – which can be appealing for those young people who struggle to build face-to-face relationships.”

But the risks of replacing human relationships with chatbots are only starting to be understood.

A study we conducted in 2023 showed university students using early chatbots for academic support copped unintended consequences, such as loneliness. Currently, the risks of social chatbots – such as dependency, displacing human friendships, withdrawing socially or building unrealistic relationships – appear to outweigh the benefits.

In recent weeks, China has put into place new rules for AI companions and chatbots, or “anthropomorphic interactive services” as they’re called in the legislation. Alongside ensuring chatbots must guard state secrets, emotional use of these tools is restricted, and expected to follow social and moral standards.

These are some of the strongest controls of social AI use introduced anywhere in the world. So what can we learn?

China’s social AI regulation

Material harms from companion AI use do occur, including in China. An 18-year-old spent 7,000 yuan (A$1,400) on customised jewellery for her AI boyfriend in their four months of “dating”.

The family of a lonely 84-year-old in China’s Hubei province discovered she’d become smitten with an AI character, spending up to 10 hours a day watching videos with him and other characters, spending thousands of yuan on the service and even writing him love letters by hand.

China’s new rules establish explicit provisions for how they expect companion chatbots to interact with humans.

The measures expect social AI bot developers to assure content does not:

• threaten national security or national honour
• violate core socialist values
• encourage self-harm or suicide
• excessively cater to emotional dependence or addiction
• provide virtual ‘partners’ or ‘relatives’ to minors
• serve unsuitable content to minors.

What’s more, users who seem overly reliant on the service must receive prominent reminders they’re speaking to a chatbot. And AI companies are required to intervene if users appear to be in extreme distress.

The risk of a ‘loneliness spiral’

As we’ve shown, there are concerns – and studies to date hint at this too – that it’s not healthy for young people to rely exclusively on AI as a trusted friend or therapist.

Lonely people experiencing social anxiety can engage in problematic use of conversational AI to alleviate feelings of loneliness and negative rumination.

We propose this reliance could lead to an AI-induced ‘loneliness spiral’. When individuals feel lonely, they’re more sensitive to feelings of social threat and disruptive emotional regulation.

So, they turn to a low-risk AI ‘friend’ for support, instead of qualified or friendly humans. However, this often doesn’t fix the initial problem of loneliness, since chatbots can actually exacerbate it.

The restrictions China has put in place for AI companions can help to break this cycle and re-prioritise human connection. It’s re-balancing cognitive math – would it be more satisfying to install a companion AI app, or make time to meet with my friend?

Lessons for Australia

Loneliness and a low sense of belonging remain major challenges in Australia, with at least one in six experiencing loneliness, with some evidence suggesting higher rates.

While China has cracked down on AI companions, Australia has been leading the charge on young people’s use of social media. eSafety has also developed codes for restricting age-related material online such as pornography, high-impact violence, self-harm, suicide and disordered eating. But they don’t address social harms of AI companions, such as emotional dependency or displacing human connection.

Prime Minister Anthony Albanese also didn’t mention social harms from AI chatbots in his major speech on AI last week.

We propose that Australia’s age-restricted material protections should extend to unintended social harm of conversations with social, analytical and functional AIs.

The Australian government needs to explore how these virtual companions interact with Australians. And we must prepare to mitigate unintended social harms before these synthetic social connections become widespread and normalised.

Australia must prioritise developing social connectedness – in workplaces, productivity strategies, education curriculum and in the design of our communities.

An authoritarian crackdown on chatbots may not be within reach for a democratic nation. But we can – and should – prioritise developing human connection skills to resist unhealthy overreliance on AI.

And Australia should leverage regulatory protections to mitigate social harms, extending on the work eSafety is already doing with social media.

Joseph Crawford and Michael Cowling

Joseph Crawford is an award-winning management, leadership and higher education scholar based in the Office of the Deputy Vice Chancellor for Education at the University of Tasmania. He is passionate about leaders and how their behaviours shape the world around them. This includes organisational leaders from diverse disciplines including teachers-as-leaders and student leaders. 

Prof Michael Cowling is an award-winning technology educator and communicator with a keen interest in mixed reality education, computer science education and information technology in society. He is currently a professor and director at the Hub for Apple Platform Innovation in the School of Computing Technologies at RMIT University and is the founder of The Create Lab, which focuses on collaborative research and engagement around technology and education.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.