The Answer Engine AI News · Sydney

Security researchers have discovered hundreds of servers running Artex, a penetration testing tool, left exposed on the open internet without proper security controls. Penetration testing tools are legitimate software used by cybersecurity professionals to test network defenses by simulating hacker attacks. However, when left publicly accessible, anyone can use these powerful tools to launch real attacks against businesses and infrastructure. The Artex AI tool automates many hacking techniques, making it easier for even less-skilled attackers to exploit vulnerabilities. The exposed servers essentially provided free access to sophisticated hacking capabilities that would normally require technical knowledge to set up and use. This represents a significant security risk, as malicious actors could have used these tools to scan for weaknesses, breach networks, or steal data from any organization. The researchers did not specify how long these servers were exposed or whether they were actively misused, but the discovery highlights ongoing problems with organizations failing to properly secure their testing tools and infrastructure.

Meta is rolling out a new feature called Spaces for its Muse AI agents. Spaces is designed as a collaborative environment where users can interact with Meta's AI helpers in new ways. The update is part of Meta's ongoing effort to add functionality and integrations to its Muse agent platform, though specific details about what Spaces actually does or how it differs from existing Muse interactions aren't provided in the available information. The announcement signals Meta's continued investment in AI assistant technology, expanding the tools available within its ecosystem. As major tech platforms compete to make their AI assistants more useful and sticky, features like Spaces represent attempts to create dedicated environments where users might spend more time engaging with AI helpers. For businesses already using Meta's platforms, this could mean new ways to leverage AI within the Meta ecosystem, though the practical applications will depend on how Spaces actually functions once fully deployed.

MIT Technology Review reports on a new humanoid robot cleaning service that's being tested in real homes. A reporter watched as a humanoid robot navigated their apartment and performed cleaning tasks, describing the experience as "a taste of the future." The article appears in MIT Tech Review's daily newsletter, The Download, which also covered Samsung's moves to capitalize on the AI boom by building infrastructure to support AI development. While the source summary is brief, it suggests these robots represent a significant step forward from industrial or warehouse robots – they're entering domestic spaces and performing household tasks that require human-like movement and decision-making. The humanoid cleaning service indicates we're moving from demonstrations and prototypes to actual commercial services using advanced robotics. This type of embodied AI – robots that can move through and interact with human environments – represents a different frontier than the language models and chatbots that have dominated recent AI news.

A team has built AutoIP, a library of machine-checked mathematical proofs for inverse problems (a type of mathematical analysis used in AI and data science). They used AI agents to write proofs in Lean, a "proof assistant" system that verifies every step of mathematical reasoning down to basic axioms. The key insight: while AI can now write formal proofs cheaply, humans still need to check that the formal statement actually says what it's supposed to say in plain language. The researchers found that the real risk isn't in the AI-generated proofs themselves—those are machine-verified—but in whether the formal statements match their natural-language descriptions. They demonstrated this by finding a published textbook theorem whose conditions were actually more restrictive than claimed. The team has organized their library so reviewers only need to check statements against their plain-language versions, never needing to read the underlying proofs. They're inviting the broader research community to build on and extend this work, suggesting this approach could help catch errors in mathematical foundations that underpin AI systems.

Researchers have developed KDASO (Knowledge-Data Dual-Driven Automated Skill Optimization), a system that helps large language models make more accurate liability decisions in areas like legal disputes, insurance claims, and ride-sharing adjudication. The problem it solves: when AI systems use pre-programmed "skills" (rule-based reasoning steps) to determine who's at fault in a dispute, those skills quickly become outdated as real-world situations change, the AI often misinterprets what the skill is supposed to do, and the growing library of skills becomes too large for the AI to process at once. KDASO addresses this by automatically learning new skills from recent cases, refining existing skills based on how the AI actually uses them (rather than how experts intended them), and intelligently selecting only the most relevant skills for each case to stay within processing limits. Testing across ride-hailing disputes, legal liability cases, and logical reasoning tasks showed KDASO significantly outperformed current general-purpose approaches. The system essentially allows AI to stay current and accurate in complex decision-making without constant manual updates from human experts.

Google is expanding access to Gemini's Business Profile management tools to Workspace accounts, which was previously limited to standard Google Business Profile users. The company has also relocated the Messages button in Google Maps and quietly removed regional restrictions from its help documentation, suggesting broader availability.

For context, Business Profile tools let you manage how your business appears across Google Search and Maps. Adding Gemini AI capabilities to Workspace means businesses using Google's paid productivity suite (Gmail, Drive, Docs for organisations) will soon be able to use AI assistance to update their profiles, respond to reviews, and manage their local presence.

The removal of region limits from Google's help pages is particularly noteworthy—it suggests these AI-powered business management features may now be rolling out globally rather than being restricted to specific countries. The moved Messages button is a smaller interface change but indicates Google is actively reorganising how businesses and customers interact through Maps.

This rollout appears to be part of Google's broader push to integrate Gemini AI across its business tools, making it easier for companies to maintain their online presence without manual updates.

The article argues that while AI visibility—how often your business appears in AI-generated answers—is real and worth tracking, most current measurement tools are unreliable. The problem isn't the concept; it's that many dashboard scores don't correlate with actual customer behavior. To make AI visibility budgets defensible, the author recommends testing whether your dashboard metrics actually match what real users experience. Instead of relying solely on vanity metrics, businesses should focus on three concrete measures: exposure (how often you appear in AI results), recall (whether users remember seeing you), and consideration (whether they'd actually choose your business based on what the AI said). The key takeaway is that AI visibility metrics need validation against real-world outcomes before you commit serious budget to improving them. Many providers offer impressive-looking scores that may not translate to customer actions.

Traditional search rankings are becoming less relevant as AI assistants and chatbots change how people find information. Search Engine Journal argues that agencies need to measure three new metrics instead: inclusion (whether your business is mentioned at all in AI responses), framing (how you're described when mentioned), and endorsement (whether the AI actively recommends you). These metrics matter more than your Google position because AI tools like ChatGPT, Perplexity, and Google's AI Overviews don't always show a ranked list—they synthesise answers and make recommendations directly. The article suggests agencies should track whether clients appear in AI responses, analyse the context and tone of those mentions, and measure how often they're positioned as the preferred option. This shift requires different reporting methods to show clients what's actually working in the AI era, rather than relying on traditional rank-tracking tools that miss the full picture of online visibility.

Australian payments company Tyro has appointed a new Chief Information Security Officer, with an initial brief to use AI to bolster the company's cybersecurity defences. The move signals Tyro's intention to leverage artificial intelligence tools as part of its security strategy, joining a growing number of Australian businesses exploring how AI can help detect threats, automate security responses, and strengthen their cyber posture. While details of the specific AI applications remain limited, the appointment reflects broader industry trends where security leaders are tasked with integrating AI capabilities into their defensive arsenals. For a payments business handling sensitive financial data, the focus on AI-enhanced security makes strategic sense — these tools can process vast amounts of security data faster than human teams and potentially identify attack patterns earlier. The appointment comes as Australian businesses face increasing cyber threats and regulatory scrutiny around data protection.