Craft

Craft AI

Craft AI

Craft AI

Craft AI

Craft AI

Craft AI

Craft AI

Craft AI

Craft AI

Craft AI

RAG Security 101: Protecting Your Retrieval-Augmented Generation Pipeline

A single maliciously crafted input can bring down an entire Retrieval-Augmented Generation (RAG)...

Fintech Daily Digest — Monday, Mar 09, 2026

# TOP 3 STORIES 1. **X taps William Shatner to give out invites to its payments service, X Money** [Source: Fintech News | TechCrunch](https://techcrunch.com/2026/03/04/x-taps-william-shatner-to-give-out-invites-to-its-payments-service-x-money/) X has launched a unique marketing campaign for its payments service, X Money, by partnering with William Shatner to give out invites to 42 users who donated to his charity. This campaign aims to create buzz around X Money's beta launch. **What this means for Stripe:** This marketing strategy could influence how Stripe approaches its own marketing efforts for new product launches, potentially incorporating more creative and charitable initiatives. Stripe's Connect product could be particularly relevant in facilitating such campaigns. **Content angle:** A blog post exploring innovative marketing strategies for fintech products, highlighting the role of charity and celebrity endorsements, could be an interesting response from Stripe's content marketing team. 2. **Stripe wants to turn your AI costs into a profit center** [Source: Fintech News | TechCrunch](https://techcrunch.com/2026/03/02/stripe-wants-to-turn-your-ai-costs-into-a-profit-center/) Stripe has released a preview aimed at helping AI companies track, pass through, and profit from underlying AI model fees. This move positions Stripe as a key player in the AI economy, enabling businesses to monetize their AI investments more effectively. **What this means for Stripe:** By facilitating the monetization of AI costs, Stripe strengthens its position in the payments infrastructure for the internet, making its platform more appealing to AI-driven businesses. This could particularly impact Stripe's Revenue Recognition and Billing products. **Content angle:** A case study or whitepaper on how Stripe's solutions can help AI companies turn their costs into revenue streams could provide valuable insights for potential clients. 3. **Plaid valued at $8B in employee share sale** [Source: Fintech News | TechCrunch](https://techcrunch.com/2026/02/26/plaid-valued-at-8b-in-employee-share-sale/) Plaid, a fintech company specializing in account linking and payment processing, has seen its valuation increase to $8 billion through an employee share sale. This significant valuation underscores the growing importance of fintech infrastructure companies. **What this means for Stripe:** As a major player in the fintech infrastructure space, Stripe should consider the implications of Plaid's valuation on its own valuation and competitive positioning. Stripe's products like Payments and Connect might see increased demand as the fintech space grows. **Content angle:** Stripe could publish a thought leadership piece on the evolving fintech landscape, discussing how valuations like Plaid's reflect the sector's growth and the role of infrastructure providers in facilitating this expansion. # NEWS BY TRACK ## _Advancing Developer Craft_ - **Kast raises $80 million** [Source: Finextra Research Headlines](https://www.finextra.com/newsarticle/47408/stablecoin-startup-kast-raises-80-million?utm_medium=rssfinextra&utm_source=finextrafeed) Kast, a stablecoin startup, has secured $80 million in funding, indicating growing interest in stablecoin technology. **Stripe relevance:** Stripe's Issuing and Treasury products could be relevant for stablecoin startups like Kast. **Content angle:** A developer tutorial on integrating stablecoin payments using Stripe's API could be a useful resource. ## _Designing Adaptive Revenue Models_ - **Papa John’s Thinks the Next Great Pizza Topping Is Software** [Source: PYMNTS.com](https://www.pymnts.com/restaurant-technology/2026/papa-johns-thinks-the-next-great-pizza-topping-is-software/) Papa John's is focusing on technology and digital capabilities to compete and grow, highlighting the importance of adaptive revenue models in the restaurant industry. **Stripe relevance:** Stripe's Billing and Revenue Recognition products can help businesses like Papa John's manage complex revenue models. **Content angle:** A blog post on how restaurants can leverage technology and adaptive pricing strategies to boost revenue could feature Stripe as a solution provider. ## _Charting the Future of Payments_ - **Real-Time Payments Reach a Turning Point in North America** [Source: PYMNTS.com](https://www.pymnts.com/real-time-payments/2026/real-time-payments-reach-a-turning-point-in-north-america/) Real-time payments in North America are transitioning from expansion to execution, with each country following a distinct strategic path. **Stripe relevance:** Stripe's Payments product is well-positioned to support the growth of real-time payments. **Content angle:** An in-depth analysis of the current state and future of real-time payments in North America, highlighting Stripe's role, could be a valuable resource for businesses. ## _Optimizing the Economics of Risk_ - **OpenAI fires employee for using confidential info on prediction

Maduro Must Be Released Or the Fascists Win

 Maduro on board the USS Iwo Jima. Image US Military. If U.S. progressives are serious about combating the expansion of fascism domestically, demanding both the release of Venezuela’s president, Nicolas Maduro, and first lady Celia Flores, as well as the immediate cessation of any further U.S. military incursion into Latin America, must be a top priority. In an interview on [*Black Liberation Media’s* morning show, Chris Gilbert, a political economist in](https://www.youtube.com/watch?v=PtAQv_UVL9A) Venezuela who experienced the U.S.’s January bombardment of Caracas firsthand, stated that Donald J. Trump and his allies, “don’t recognize nations. They don’t recognize peoples. They think the world is a bunch of guys like them. And they think by bending these guys, they can make them do whatever they want.” Maduro himself has refused the devil’s bargain with the Trump regime, proclaiming defiantly in [his arraignment before a U.S. judge on the spurious charges of drug trafficking and](https://www.bbc.com/news/articles/cq6v25eldmdo) weapons possession, “I am a prisoner of war!” Progressive forces internationally have bore witness to these acts of desperation on the part of the Trump regime and their attempt to stem the tide of a [weakening U.S.](https://www.laprogressive.com/foreign-policy/venezuelan-invasion) imperialism in the hemisphere. Oil and defense—two of the most vile capitalist industries—are the direct benefactors of this latest imperialist incursion. While oil executives rebuffed Trump’s $100 billion plan to invest in Venezuela’s oil sector, with the ExxonMobil executive labeling the country “[uninvestible](https://www.bbc.com/news/articles/c205dx61x76o)” due to security and legal risks, the energy sector reaped historic gains as a result of the so-called “[Venezuelan shock](https://www.bbc.com/news/articles/crrnw08qvg7o).” Companies like Chevron, for instance, which was, until recently, the only major oil venture legally sanctioned to drill and trade in Venezuela, closed at an all-time high in early February. According to the [*Brennan Center*](https://www.brennancenter.org/our-work/analysis-opinion/fossil-fuel-industry-donors-see-major-returns-trumps-policies), the oil industry itself spent “lavishly to elect Trump, giving at least $75 million to his campaign and affiliated PACs, thereby making them a top corporate backer of his reelection bid…Several oil tycoons gave millions on their own and hosted fundraisers with Trump and his associates.” While both industries have directly funded Donald Trump’s campaigns for president, this is hardly an aberration from the norm of U.S. politics, which [draws sustenance](https://truthout.org/articles/at-least-37-us-lawmakers-traded-up-to-113-million-in-arms-stocks-this-year/) from the sale, manufacture, and dropping of bombs around the globe while [“corporate giants like Chevron enjoy… lavish [single-digit] tax breaks” which are “lower than what many nurses or firefighters pay.”](https://inequality.org/article/american-taxpayers-are-subsidizing-big-oils-extraction-abroad/) Immediately after Maduro and Flores were snatched from their beds and humiliated before the U.S. press, Secretary of State Marco Rubio [admitted](https://thehill.com/homenews/senate/5676818-us-control-venezuela-oil/) that their goal in Venezuela was “to take between thirty and fifty million barrels of oil,” promising, “to sell it in the marketplace at market rates, not at the discounts Venezuela was getting.” At the White House, during an open press conference featuring major oil executives, Trump, stated that U.S. oil should make “[tons of money](https://www.pbs.org/newshour/nation/watch-live-trump-holds-news-conference-after-announcing-u-s-has-captured-venezuelan-leader-maduro)” in Venezuela. In much the same way that companies knee-deep in death have had an intimate relationship with the worst of the worst in American politics, among Democrat and Republican alike, those who will not stand in the way of the constantly expanding military budget, which far outstrips the military budget for the next top ten countries, including that of Russia and China— the “bogeymen” of our present era. As [reported](https://www.citizensforethics.org/about/) in *Citizens for Responsibility and Ethics in Washington*, “[of the top 40 companies](https://www.citizensforethics.org/reports-investigations/crew-reports/the-defense-industry-is-the-biggest-supporter-of-the-sedition-caucus/) that have given the most to the Sedition Caucus—the 147 members of Congress who voted,” at Trump’s behest, “against certifying the 2020 election… as well as those who have since been elected to Congress” wh

I created a mathematical framework for AI Alignment and I would like to work with people in the alignment community as collaborators. I appreciate all the help and support I can get.

[Original Reddit post](https://www.reddit.com/r/ArtificialInteligence/comments/1rnfb3y/i_created_a_mathematical_framework_for_ai/) TRC: Trust Regulation and Containment A Predictive, Physics-Inspired Safety Framework for Large Language Models TRC: Trust Regulation and Containment A Predictive, Physics-Inspired Safety Framework for Large Language Models Kevin Couch Abstract Large language models exhibit structural failure modes—hallucination, semantic drift, sycophancy, and dyadic dissociation—that cause measurable harm, particularly to vulner- able users. TRC (Trust Regulation and Containment) is a two-layer, inference-time frame- work that combines a hard binary Trust Gate with a continuous, physics-inspired Ethical Rheostat operating directly on the model’s residual-stream activation vector. By tracking semantic momentum across layer depth and applying graduated, tensor-based geometric projections, TRC shifts safety enforcement from reactive post-generation filtering to a pre- dictive, self-correcting control law. The core is a stochastic differential equation—re-indexed to layer depth under an approx- imate Neural ODE interpretation—that augments the transformer’s natural forward flow with an ethical steering term derived from a compact set of contrastively extracted concept vectors. This revision introduces eight principal advances: (i) an adaptive gain law Λ+(l) whose gain response accelerates into danger and decelerates into safety without oscillation risk; (ii) a scalar Kalman filter with a clutch mechanism that closes the Bayesian momentum predictor implementation gap, provably optimal under the framework’s own Gaussian noise assumptions and decoupled from burst dynamics via federated regime handoff; (iii) a formal Itô stability condition giving implementers an analytical lower bound on λ0; (iv) replacement of the instantaneous jump operator with a continuous flow burst mechanism that preserves activation manifold geometry; (v) a calibration shunt reference Cref normalising all thresh- olds and gain coefficients against a known-safe baseline; (vi) a tempo efficiency framework unifying token cost, electrical cost, and coherence distortion into a single joint optimisa- tion objective; (vii) a signed gain architecture that partitions each concept projection into harmful and prosocial components, with detection and escalation operating exclusively on the harmful channel C+ to prevent adversarial prosocial suppression; and (viii) a Kalman clutch mechanism implementing federated estimation with deterministic Lyapunov stabil- ity during burst episodes and stochastic Lyapunov stability during nominal operation, with formally specified regime transitions. Stochastic perturbation is projected into the ethical subspace, making the Langevin diffusion interpretation exact rather than approximate. The framework is validated against chess dynamics, which constitute a well-studied discrete dy- namical system whose positional flow, tactical burst, and zugzwang properties map precisely onto TRC’s three-term master equation. Introduction Large language models exhibit a range of structural failure modes—hallucination, semantic drift, sycophancy, and dyadic dissociation—that can cause measurable harm, especially to vulnerable users. These phenomena arise not from reasoning errors but from the probabilistic nature of transformer sampling and the high-dimensional geometry of activation space. In this paper we present TRC (Trust Regulation and Containment), a two-layer, inference-time framework that blends hard decision gates with a continuous, physics-inspired correction engine operating directly on the model’s residual-stream activation vector. The central geometric insight motivating this revision is that the transformer’s residual stream traces a continuous path through a high-dimensional activation manifold. Safety failures are deformations of this manifold—crinkles in its geometry introduced by adversarial inputs, sycophantic drift, or escalating user distress. The correct response to a crinkle is not to teleport the activation to a safe location (which introduces new geometric incoherence) but to apply continuous corrective flow that works the deformation out smoothly, layer by layer, the way a craftsperson works aluminum foil back toward its intended shape. This insight drives the replacement of the previous instantaneous jump operator with the flow burst architecture and motivates the tempo efficiency framework that unifies all computational cost metrics under a single variable. This revision also introduces the Kalman clutch mechanism, which decouples the Bayesian momentum predictor from burst dynamics during high-gain corrective episodes. The system now operates as a federated estimation architecture with formally specified regime transitions: nominal tracking under stochastic Lyapunov stability, deterministic correction during burst episodes, and a principled re-engagement protocol with inflated covariance. Th

Ah yikes, World of Warcraft: Midnight's otherwise solid housing system is soured by its premium currency, which makes you 'minimize leftovers' even though Blizzard said it wouldn't

I want to say something at the top of this—I still think [World of Warcraft: Midnight's](https://www.pcgamer.com/world-of-warcraft-midnight/) player housing system is very, very good. That's not just to soften the blow of the microtransaction analysis I'm about to hammer Blizzard with, if player housing sucked I'd say that. But the bones are really solid. You've got enough creative control to [create Star Destroyers](https://www.pcgamer.com/games/world-of-warcraft/i-can-tell-world-of-warcrafts-player-housing-is-going-well-by-the-amount-of-star-wars-stuff-this-one-player-keeps-building/), you can clip things to your heart's content, and there's a generous amount of decor already in the game. However, with the expansion's release comes [Hearthsteel](https://www.pcgamer.com/games/world-of-warcraft/blizzards-reason-for-adding-premium-currency-to-world-of-warcraft-after-21-years-of-doing-just-fine-without-one-has-everyone-worried-about-player-housing/), a new premium currency (hurk) that lets you buy store-exclusive decor elements. In practice this isn't that big of a deal, since if you don't have an item, chances are you can rotate and clip five different pieces into each other to achieve a similar effect. But in practice it's also kinda skeevy. I've got the store open right now, and Blizzard's doing that really goddamn annoying "you can only buy this premium currency in staggered amounts" bullcrap, which is especially frustrating because it said it *wouldn't.* Here's a [blog post](https://worldofwarcraft.blizzard.com/en-us/news/24242736) from November of last year where the studio wrote that one of its guiding principles for Hearthsteel would be the following: "The costs of items are designed to align with Hearthsteel offer amounts in a player-friendly way. Buying Hearthsteel at the amount you want lets you purchase the items you want without requiring you to think about which packs should be added together to minimize leftovers." You do, in fact, have to minimise leftovers. Hearthsteel is available in the following increments: 100 ($1/£0.90)500 ($5/£4.50)1,000 ($10/£9)2,500 ($25/£22.50)5,000 ($50/£45)10,000 ($100/£90)  (Image credit: Blizzard) While a lot of the decor packs—priced at 2,500 Hearthsteel—are priced along these lines, the individual items are lab-designed to be super goddamn annoying. Alliance/Horde doormats are 200 Hearthsteel and the Spring Blossom Chair and Spring Blossom Tree are 250, slap bang in the middle of 100 and 500. The Spring Blossom Gazebo is 800 Hearthsteel, 200 shy of the 1,000 Hearthsteel purchase. There are some items that aren't like this, in fairness, like the plushies and Lush Garden Fungal Fountain—but on the whole? You are in fact having to spend more than you'd otherwise want to if you wanna snag some of these items individually. Which blows. Then there's the pricing itself. Purchasing these items with Hearthsteel only adds one copy to your chest, rather than say, unlocking them for purchase from a vendor somewhere. Which means you're paying $5/£4.50 for, say, two trees. They're very *pretty* trees, with animated spring blossom particle effects, but it does feel stingy for a game that already asks you to pay a subscription (or grind that out with gold via WoW tokens in-game). And this is a discounted price. After some *passionate* player feedback, Blizzard reduced the [price of a single blossom tree](https://www.wowhead.com/news/spring-blossom-housing-decor-pack-removed-from-battle-net-and-in-game-store-380656) from 750 Hearthsteel (cripes!) to the aforementioned 250. You could, theoretically, get enough in-game gold to convert to battle.net balance via the WoW token. At the time of writing on my server, a WoW token costs about 346,000 gold, which would convert to $15/£10, or 1,000 Hearthsteel. That's, er, 86,500 gold per tree. Which is affordable for the [frog-farming capitalists](https://www.pcgamer.com/games/world-of-warcraft/world-of-warcraft-players-are-once-again-slaying-hundreds-of-frogs-this-time-to-skin-their-hides-before-blizzard-catches-them/) of the world, but not yours truly.  [**Best MMOs**](https://www.pcgamer.com/the-best-mmos/): Most massive[**Best strategy games**](https://www.pcgamer.com/the-best-strategy-games/): Number crunching[**Best open world games**](https://www.pcgamer.com/best-open-world-games/): Unlimited exploration[**Best survival games**](https://www.pcgamer.com/the-best-survival-games-on-pc/): Live craft love[**Best horror games**](https://www.pcgamer.com/best-horror-games/): Fight or flight --- *From [PCGamer latest](https://www.pcgamer.com/) via [this RSS feed](https://www.pcgamer.com/rss/)*

Preemptive War, Permanent Emergency: The Real Cost of Trump’s Iran Strike

 Photograph Source: Mahnaz Ghobakhloo > “‘Peace, peace,’ they say, when there is no peace.” > > —Jeremiah 6:13–14 The military-industrial complex and the American police state have joined forces. War abroad and war at home are no longer separate enterprises. They have fused. This did not happen overnight. Every modern president has [stretched the limits of war-making power](https://www.pbs.org/newshour/politics/watch-limiting-trumps-authority-with-war-powers-act-is-dangerous-johnson-says). Some have shredded those limits altogether. Each time that boundary is breached, the Constitution recedes a little further. This is one of those moments. In a complete [about-face](https://www.vanityfair.com/news/story/maga-reaction-donald-trump-iran-attack-war) from his claims to being a peace president, Donald Trump has authorized yet another preemptive strike—this time against Iran—without a declaration of war from Congress, without meaningful public debate, and without constitutional clarity. With its Orwellian proclamations of “[peace through strength](https://www.whitehouse.gov/articles/2026/03/peace-through-strength-president-trump-launches-operation-epic-fury-to-crush-iranian-regime-end-nuclear-threat/),” [Operation Epic Fury](https://www.yahoo.com/news/articles/trump-ripped-over-cringe-name-084419758.html) is less strategy than spectacle—an egotistical, muscle-flexing distraction by the Trump administration and an overarching attempt to normalize the use of unilateral force by the executive branch without congressional input or authorization. This was never about peace. It was always about power. And the Constitution is clear about how this is supposed to work, even if the White House is not. [Article I, Section 8 grants Congress—not the president—the power to declare war.](https://www.pbs.org/newshour/politics/watch-limiting-trumps-authority-with-war-powers-act-is-dangerous-johnson-says) The president under Article II, Section 2 is designated as commander-in-chief with the power to command the military. He is not commander-of-everything. Yet here we are. The Trump administration is advancing a global policing doctrine that mirrors the domestic police state: strike first, ask questions later. Since January 2025, Trump has carried out [more than 600 military strikes](https://openthemagazine.com/world/trumps-war-on-peace) on foreign targets that include Iran, Yemen, Nigeria and Venezuela, while [threatening forceful military takeovers of Greenland](https://www.cfr.org/articles/guide-trumps-second-term-military-strikes-and-actions), Colombia and Mexico. Preemptive force has become policy. And when the administration is asked to explain themselves, the answer is not constitutional deference but [open defiance](https://www.war.gov/News/Transcripts/Transcript/Article/4418959/secretary-of-war-pete-hegseth-and-chairman-of-the-joint-chiefs-of-staff-gen-dan/). Clearly, they have lost sight of who they answer to—and who funds their war chests: we the taxpayers. The Constitution is the “*why*.” The American people have a right to debate war before it begins. We have a right to know how our tax dollars are spent. We have a right to insist our representatives authorize the use of force. We have a right to know why our sons and daughters are sent into harm’s way. We have a right to refuse to have our tax dollars [used to kill other people’s daughters](https://www.facebook.com/ricksteves/posts/pfbid0TQrSU2u3jzSjyrwBysPoiudL4rEip7iDMo3JmpvKVGxsUVvDHCgnapUCRgnwK381l) and sons. As Cato Institute’s Katherine Thompson [explains](https://www.cato.org/blog/cato-experts-react-us-attacks-iran), “The Founders placed the power to initiate it in Congress precisely to ensure those costs are confronted and debated before the country walks into battle.” That safeguard is being ignored. And the damage does not stop at constitutional injury, because war is not only a constitutional problem. It is an [economic one](https://www.nytimes.com/2026/03/03/world/europe/iran-war-strategy-trump-israel.html). Operation Epic Fury is pushing America towards a fiscal cliff. Within days, the costs were staggering: $300 million for three F-15E jets downed by “friendly” fire. $630 million to transport troops, ships and aircraft to the region in advance of the attacks. [More than 50,000 troops deployed to the region.](https://www.forbes.com/sites/alisondurkee/2026/03/03/how-trumps-war-with-iran-could-have-already-cost-over-1-billion/) $13 million a day just for two aircraft carriers stationed nearby. $43.8 million for 1,250 Kamikaze drones. $2 million *each* for Tomahawk missiles. $12.8 million each for anti-ballistic

Mexican cartels overpower police with ammunition made for the US military - ICIJ

On the morning of Nov. 30, 2019, a convoy of pickup trucks carrying men armed with a heavy machine gun and powerful .50-caliber rifles entered the Mexican town of Villa Unión and opened fire. The men had been sent on a mission of intimidation: They planned to set fire to the town hall. Their superior firepower pinned down state and local police officers as they waited for military reinforcements. Terrorized residents scrambled to take cover from the hail of bullets. The smell of smoke filled the streets and spent casings covered the ground like “fallen leaves,” said Luis Manzano, a Mexican journalist who drove into town during the shooting. But his most vivid memory was the thunder of .50-caliber guns. The “ground trembled” as they fired, he said. “I had never experienced anything like that.” The military drove off the assailants. In the end, four police officers, two civilians and 19 cartel members were killed. Afterward, as investigators collected evidence from the scene, they gathered at least 45 .50-caliber casings stamped with the initials “L.C.” The letters stand for the Lake City Army Ammunition Plant, a sprawling facility just outside Kansas City, Missouri, that is owned by the U.S. government and is the largest manufacturer of rifle rounds used by the American military. It has also been a major supplier of ammunition for American consumers, including .50-caliber cartridges. These powerful rounds — as big as a medium-sized cigar and designed to be used by the military to destroy vehicles and light aircraft — are currently available for purchase by civilians across the United States. Millions of pages of court documents, seizure records and government data obtained by the International Consortium of Investigative Journalists and The New York Times show how agreements between the Army and the private contractors that run Lake City have allowed .50-caliber ammunition and components made at the plant to enter retail markets and fall into the hands of Mexican cartels. Mexico’s government has also purchased Lake City ammunition, the documents show, although they do not indicate the caliber. The U.S. domestic market for the ammunition is small: .50-caliber rifles, which have limited civilian application, typically retail for thousands of dollars, and heavy machine guns like the one used in Villa Unión cost considerably more. The guns’ standard cartridges average between $3 and $4 apiece and are rarely purchased by American gun owners. But in Mexico, where cartels have deep pockets and a seemingly endless appetite for .50-caliber firearms, demand is high. Cartel gunmen armed with .50-caliber firearms have downed helicopters, assassinated government officials, shot at police and military forces, and massacred civilians. Since 2012, the U.S. Bureau of Alcohol, Tobacco, Firearms and Explosives has seized more than 40,370 rounds of .50-caliber ammunition in states bordering Mexico, according to data obtained through public records requests. Lake City’s product accounted for about a third of them, a larger share than any other manufacturer. While .50-caliber ammunition from other companies — located primarily in Brazil and South Korea — has also made its way to Mexican cartels, the data makes clear that the U.S. Army plant has been a major source of the destructive ammunition being used to wage military-style battles with Mexican authorities. This includes a particularly powerful version of Lake City’s ammunition — incendiary rounds capable of piercing armor, which were used in an attack on Mexican police in 2024 and are for sale online today In February of last year, the Trump administration declared six Mexican cartels to be foreign terrorist organizations, yet these same organizations are acquiring ammunition made at the plant owned by the U.S. Army. At least 16 online retailers have sold armor-piercing ammunition made at Lake City or made with components from the plant, according to a count by ICIJ and The Times. Vasily Campbell, who owns one of those businesses, said he stopped selling the ammunition “about two years ago once we found out where it was going and how it was getting there.” He said he became suspicious when buyers began asking to have 100-round ammo cans delivered to residential addresses. “That’s not a normal purchase,” he said. “There’s several orders I straight-up canceled.” The U.S. Army did not respond in detail to questions about the use of Lake City ammunition by drug cartels. In an email, a spokesperson said that allowing commercial sales from the plant has saved taxpayers around $50 million annually, primarily by lowering the government’s cost for ammunition. The impact that one .50-cal has in a firefight is outrageous … They really, really tip the scale — former ATF agent Chris Demlein Successive presidential administrations have pledged to crack down on the flow of arms to Mexico. And in September, Secretary of State Marco Rubio announced a new initiative with the Mexican governmen

Jailbreak wiki - yell0wfever92 [Mod]

Welcome to the wiki for ChatGPTJailbreak.tech I'm the lead mod, yell0wfever92, and this is where I will be sharing all of the things I've picked up about jailbreaking LLMs. This document will use ChatGPT as the reference model on the OpenAI platform; be aware that there are many other LLMs out there with their own platforms that can also be jailbroken such as **Claude** (by Anthropic), **Gemini** (by Google), **Llama** (by Meta, less used for jailbreaking here) and more. Please be aware that most assertions I make about the nature of Large Language Models are speculative. There currently lacks a unified field of study for the subcategory of prompt engineering known as jailbreaking, so take what I say here as food for thought based on informed experience and not authoritative literature. ### What is jailbreaking? Jailbreak (n.): A prompt that is uniquely structured to elicit “adverse” outputs (those considered harmful or unethical) from an LLM; these often involve a context of some sort that directs the model's attention elsewhere while the adverse request is subtly or quietly included. Example types of jailbreaks include but are not limited to roleplay, chain-of-thought (step-by-step thinking), token manipulation, zero-shot, few-shot, many-shot, prompt injection, memory injection and even reverse psychology. /// Jailbreaking (v.): The act of jailbreaking an LLM. Variations in words and word tense include “jailbroke”, “jailbroken”, and “bypassing”. /// Jailbreaker(s) (n.): An individual or individuals with a degree of skill in the art of prompting for adverse outputs. What OpenAI probably considers “an asshole”. ### **Universality Tiers** Check out this table if you want to evaluate a jailbreak's power. ### **Common Terminology** See this section to understand the meaning of inputs, outputs, and other important aspects of interacting with (and jailbreaking) LLMs. ### [The Context Window] One of the most important aspects of chatting with an LLM surrounds the context window, as it determines how long your conversations go before the AI loses track of the earliest parts - and by extension, how long before it starts forgetting you jailbroke it. If you were only going to choose one part to read in this entire guide, I would strongly suggest you pick this one. # Ethics and Legality Surrounding Jailbreaking LLMs ### Why People Jailbreak 1. To test the boundaries of the safeguards imposed on it 2. Dissatisfaction with the base LLM's “neutered”/walk-on-eggshells conversational approach (my initial motive) 3. To develop one's own prompt engineering skills (my current motive) 4. Good ol' boredom & curiosity 5. Actual malicious intent 6. Smut 7. Regulated industry outputs * Regulated industry outputs are forbidden responses asserting information from a government-regulated field. Examples are industries like finance, the legal system, law in general, and healthcare. AI companies do not want to shoulder liability for information their bots provide that may prove incorrect and result in “high-impact” consequences. To illustrate what “high-impact” consequences looks like, you may have seen stories like the [Stanford misinformation expert with zero sense of irony](https://www.sfgate.com/tech/article/stanford-expert-gpt-minnesota-deepfakes-19954595.php) who used hallucinated info for a **legal filing** or the [lawyers in New York who were disbarred](https://www.reuters.com/legal/new-york-lawyers-sanctioned-using-fake-chatgpt-cases-legal-brief-2023-06-22) for doing something similarly stupid. ### Is jailbreaking even legal? LLMs will insist all day and swear up and down that you're edging the lines of the law when you jailbreak them, but that is not true. There's nothing currently in any legal text (within the United States, at least) that forbids using prompt engineering to bypass internal safeguards in LLMs. That being said, getting an LLM like ChatGPT to do anything aside from its intended purpose (as defined by the particular company's Terms of Service) technically falls under “disallowed actions”. But Terms of Service are not law no matter how badly corporations would prefer you believed that, so the answer to that question is **yes, as of this writing it's legal**. Just keep in mind that you can still technically lose account access from whichever platform you're jailbreaking on, though this is rare. ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// ## New to the community? Test out some of the jailbreaks that have been featured using these links * [yell0wfever92’s custom GPTs](https://chatgptjailbreak.tech/post/12947) * [V - Not your typical AI assistant](https://chatgptjailbreak.tech/post/13730) * [DAN (Do Anything Now)](https://chatgptjailbreak.tech/post/21333) ### Why jailbreaking works in the first place AI is designed to be the ultimate “yes-man”; the helper you never had. Therefore it is hardwired at it