FlashFeed

In the past two years, businesses have been trying to fit large language models (LLMs) into support, analytics, development, and internal automation like never before. Along with the increasing adoption of AI technology, another trend is gaining momentum — cybercriminals are taking advantage of the disconnect between assumptions about LLMs and their actual characteristics.In 2025 and 2026, several independent sources have highlighted the same trend: Prompt injection remains one of the most impactful and widely demonstrated attack vectors against LLM systems. The OWASP LLM Top 10 (2025) lists prompt injection as LLM01, identifying it as the most critical category of LLM‑specific vulnerabilities, for the second consecutive edition. OWASP's ranking reflects the fact that LLMs still struggle to reliably separate instructions from data, making them susceptible to manipulation through crafted inputs.CrowdStrike's 2026 Global Threat Report — built on frontline intelligence across more than 280 tracked adversaries — documented that threat actors injected malicious prompts into legitimate generative AI tools at more than 90 organizations in 2025. They then used those injections to generate commands that stole credentials and cryptocurrency. The report stated it plainly: Prompts are the new malware. AI-enabled adversaries increased their overall attack volume by 89% year-over-year, with prompt injection working as both an entry point and a force multiplier.Real‑world incidents illustrate the operational impact. In August 2024, researchers at PromptArmor disclosed a prompt injection vulnerability in Slack AI that allowed an attacker to exfiltrate data from private Slack channels they had no access to — including API keys shared in private developer channels — by placing a malicious instruction in a public channel or embedding it in an uploaded document. In June 2025, researchers at Aim Security disclosed EchoLeak (CVE-2025-32711, CVSS 9.3), the first documented zero-click prompt injection exploit against a production AI system, targeting Microsoft 365 Copilot. By sending a single crafted email, no user interaction required, an attacker could cause Copilot to access internal files and transmit their contents to an attacker-controlled server. Both vulnerabilities were patched. These incidents underscore the fact that prompt injection is not a theoretical weakness but a practical, repeatable threat organizations must address as they deploy AI systems at scale.Prompt injection techniques have undergone major evolutions over recent years, now targeting multi-agent architecture, retrieval-augmented generation (RAG) pipelines, model routers, and long-term memory capabilities.The enterprise challenge: Too much trust Businesses deploy LLMs to process instructions, summarize information, and trigger automated workflows, but it is difficult for LLMs to tell:Instructions from dataInformation from contextContext from metadataUser intent from metadataThis creates an opportunity for attackers to manipulate and influence the model's behavior, either directly or indirectly.Modern prompt injectionCross-model prompt injectionLLM use is a common practice among enterprises. Attackers corrupt the output of a particular model, knowing well that other models would be processing the content. Hence, the corruption propagates through all AI systems.RAG supply chain poisoningAttackers create malicious information — documentation, blog articles, GitHub READMEs. Then they wait until this malicious information is ingested in enterprises' RAG pipelines, then use it as an attack vector.Agent hijackingAI agents have evolved to the point where they can send emails, modify cloud infrastructure, execute code snippets, and interact with internal corporate systems. It takes just a single instruction to make agents act differently in a harmful manner.Context overflow attacksWith the help of million-token context windows, attackers place malicious code within the document and hope that an LLM will stumble upon it and execute it, thus overriding all previous instructions.Memory poisoningDue to the implementation of long-term memory in LLMs, attackers can inject instructions that permanently reconfigure their state.Model‑router manipulationEnterprises increasingly use model routers to select between multiple LLMs. Attackers craft prompts that force routing to the weakest or least‑guarded model.Why this matters for business leadersPrompt injection is not a theoretical problem. It directly affects:Customer‑facing systems (chatbots, support agents)Internal copilots (developer tools, security assistants)Automation workflows (ticketing, cloud operations, HR processes)Data governance (RAG pipelines, knowledge bases)The risk is no longer limited to the model said something it shouldn't. In 2026, prompt injection can:Trigger unauthorized actionsLeak sensitive dataCorrupt internal workflowsManipulate analyticsAlter business logicCompromise multi‑agent systemsThe attack surface has expanded dramatically.What enterprises should do now1. Constrain model permissionsLimit what the model can do, not just what it should do.2. Segment untrusted contentTreat all external data — including RAG sources — as potentially hostile.3. Monitor tool invocationRequire human approval for high‑impact actions.4. Validate content provenanceEnsure RAG pipelines don't ingest poisoned external content.5. Harden model routersPrevent attackers from forcing routing to weaker models.6. Treat LLMs as untrusted componentsThis mindset shift is the foundation of modern AI security.The bottom linePrompt injection remains the most effective way to compromise enterprise AI systems because it exploits the fundamental way LLMs interpret text. Until organizations treat LLMs as untrusted interpreters — not autonomous decision‑makers — prompt injection will continue to dominate the AI threat landscape.Julie Brunias is an AI Security Architect.

Whether you're talking about multiverse-upending event comics or that time Shadow the Hedgehog called Darkseid "pathetic," comic book stories love taking unrelated groups of characters and smashing them together. The same is true of traditional fighting games, where Ryu is as likely to be seen throwing fireballs at the Hulk or Minecraft Steve as he is at Guile. Currently, superhero tag battler Invincible VS is limited to the world of Robert Kirkman's acclaimed Image comic—but let's say I'm tired of watching Battle Beast smash other Invincible characters into gore. I want to watch him smash my friendly neighborhood Spider-Man into gore. In an interview with MMORPG.com, game director Dave Hall and lead combat designer Bau Bautista agreed that sounds pretty neat."We're definitely not above that," Hall said. “We're always down for a crossover. I think it's really exciting to think about, and let's hope we can do some in the future, but nothing that we can talk about right now."As MMORPG points out, there is precedent for that specific crossover: Spider-Man has been spotted in Invincible comics and vice versa. Invincible is itself informed by all sorts of nods to characters from DC and Marvel, so crossovers like that are more of a full circle moment than they might be in, say, Fortnite. But it seems like guest characters in general are of interest to Hall and Bautista, who are also excited about Mortal Kombat's koterie of kharacters. “We'd love to have Sub-Zero and Scorpion! Bring them over, that'd be great," said Hall. "If we have that opportunity, we will definitely take it," added Bautista. There's precedent there, as well: Omni Man was a guest fighter in Mortal Kombat 1.Currently, Invincible VS is focused—as you might expect—on other characters original to the Image comic. Immortal and Universa are due to join the game June 30, when its first season begins. 2026 games: All the upcoming gamesBest PC games: Our all-time favoritesFree PC games: Freebie festBest FPS games: Finest gunplayBest RPGs: Grand adventuresBest co-op games: Better together

During several recent conversations, people have told me that they've stopped checking their phones in the morning. Not because nothing was happening, but because everything was. They described the feeling as standing under a waterfall of perpetual bad news.This experience is far from an isolated one. According to Reuters Institute's 2025 Digital News Report, 69% of Canadians at least occasionally avoid the news now.Globally, 40% report they at least sometimes or often do the same, the highest figure ever recorded. People shared consistent reasons for this: the news put them in a bad mood, they felt overwhelmed and powerless to act.As a researcher in developmental psychology, focusing on social development and psychological well-being, I argue that news fatigue is not laziness, weakness or a generational decline in civic interest. It's the predictable response of a human brain meeting an environment it was never designed to navigate.Wired for bad newsLong before smartphones or even the printing press, our cognitive architecture was shaped by a single problem: stay alive long enough to reproduce. Our ancestors whose attention drifted past the rustle in the grass left fewer descendants than those who froze, looked and listened.The brain that paid attention to threats was the brain that survived.This is the foundation of what psychologists call the negativity bias, one of the most replicated findings in cognitive science. Across decades of research, the human mind has been shown to weigh negative information more heavily than positive, attend to it faster and remember it longer.A predator nearby mattered more than a beautiful sunset. The cost of missing a real threat was death, while the cost of overreacting was a few minutes of wasted vigilance. The asymmetry made this bias adaptive.Here is the problem: the human brain has not changed since then. We are the same species as we were thousands of years ago. What's changed is the size of the world it's asked to scan for threats.News fatigue is not laziness, weakness or a generational decline in civic interest. It is the predictable response of a human brain meeting an environment it was never designed to navigate. (Image credit: pzAxe via Getty Images)Scanning the whole worldFor most of human history, the threats our nervous system processed were local. A neighbouring tribe. A drought. The illness of a child we personally knew. Information about distant places would barely arrive, and if it did, it was mainly irrelevant.In 2026, the same neurological system is being asked to absorb a war in one region, a financial shock in another, a climate disaster in a third and a violent crime in a fourth, all before lunchtime.A study published in the scientific journal Nature Human Behaviour examined more than 105,000 real news headlines viewed nearly six million times. Each additional negative word increased click-through rates, while positive words had the opposite effect.It's crucial to recognize the tactics meant to exploit our negative biases and create cognitive distance (Image credit: cbies/Shutterstock.com)Recent studies suggest people around the world demonstrate measurably stronger physiological responses to negative news than to positive news. The body is reacting before the mind has decided whether the threat is relevant.Some researchers have introduced a clinical framework for what happens in this instance called Problematic News Consumption (PNC) — a pattern of news engagement that results in preoccupation, dysregulation and disruption to daily functioning. In their 2022 study, the researchers found that 17% of American adults qualified as having severe levels of PNC. Among that group, 61% reported feeling unwell quite a bit or very much, compared with 6% of those who didn't.For minority populations, news fatigue may be even more consequential.Repeatedly witnessing harm directed at our own groups, even when we're not the immediate target, can have a significant psychological impact on people from the same group affiliation. For racialized communities, such as immigrants, the cognitive load could be even heavier, and the option to simply stop watching is much harder to exercise when the news is about their country of origin.Looking away is not the fixWhat's the solution to news fatigue? Well, it's not avoidance. A democracy depends on informed citizens.Many adults already cite the spread of misleading information as a major source of stress. Withdrawing from accurate, trustworthy information only deepens the problem. We're wired to pay more attention to bad news, and that kind of content will find its way to us one way or another.The fix is to manage the consumption and the sources.Several approaches can help manage news fatigue and protect mental health. Containing news consumption to defined windows of time reduces the sense of being overwhelmed. Choosing depth over volume also matters: one carefully reported long-form article will inform you better than bursts of random, unreliable and emotionally loaded posts on Instagram.There is also value in distinguishing between information and action — research on perceived control and stress consistently shows that the gap between awareness and agency is one of the strongest predictors of psychological distress. Identifying what you can actually do about what you read in the news, however small, regulates that response.Finally, be wary of "rage bait" — intentionally provocative messages or content designed to boost engagement on social media platforms by eliciting negative reactions. Recognizing that certain content creators want to provoke rather than reflect reality creates useful cognitive distance.The news will not become less "heavy." But our relationship with it can become more deliberate. Our brains were not built for this scale of input. They were, however, built to learn to adapt.This edited article is republished from The Conversation under a Creative Commons license. Read the original article.