While state-of-the-art diffusion models (DMs) excel in image generation, concerns regarding their security persist. Earlier research highlighted DMs' vulnerability to data poisoning attacks, but these studies placed stricter requirements than conventional methods like'BadNets' in image classification. This is because the art necessitates modifications to the diffusion training and sampling procedures. Unlike the prior work, we investigate whether BadNets-like data poisoning methods can directly degrade the generation by DMs. In other words, if only the training dataset is contaminated (without manipulating the diffusion process), how will this affect the performance of learned DMs?

With ChatGPT and Bard both becoming more and more popular, many students are being tempted to use AI chatbots to cheat on their essays. But one teacher has come up with a clever trick dubbed the'Trojan Horse' to catch them out. In a TikTok video, Daina Petronis, an English language teacher from Toronto, shows how she can easily spot AI essays. By putting a hidden prompt into her assignments, Ms Petronis tricks the AI into including unusual words which she can quickly find. 'Since no plagiarism detector is 100% accurate, this method is one of the few ways we can locate concrete evidence and extend our help to students who need guidance with AI,' Ms Petronis said.

While state-of-the-art diffusion models (DMs) excel in image generation, concerns regarding their security persist. Earlier research highlighted DMs' vulnerability to backdoor attacks, but these studies placed stricter requirements than conventional methods like 'BadNets' in image classification. This is because the former necessitates modifications to the diffusion sampling and training procedures. Unlike the prior work, we investigate whether generating backdoor attacks in DMs can be as simple as BadNets, i.e., by only contaminating the training dataset without tampering the original diffusion process. In this more realistic backdoor setting, we uncover bilateral backdoor effects that not only serve an adversarial purpose (compromising the functionality of DMs) but also offer a defensive advantage (which can be leveraged for backdoor defense). Specifically, we find that a BadNets-like backdoor attack remains effective in DMs for producing incorrect images (misaligned with the intended text conditions), and thereby yielding incorrect predictions when DMs are used as classifiers. Meanwhile, backdoored DMs exhibit an increased ratio of backdoor triggers, a phenomenon we refer to as `trigger amplification', among the generated images. We show that this latter insight can be used to enhance the detection of backdoor-poisoned training data. Even under a low backdoor poisoning ratio, studying the backdoor effects of DMs is also valuable for designing anti-backdoor image classifiers. Last but not least, we establish a meaningful linkage between backdoor attacks and the phenomenon of data replications by exploring DMs' inherent data memorization tendencies. The codes of our work are available at https://github.com/OPTML-Group/BiBadDiff.

In Stanley Kubrick's masterful film 2001: A Space Odyssey, the supercomputer HAL 9000 (heuristically programmed algorithmic computer) uses artificial intelligence to detect emotion and suffering, and controls all of a spaceship's systems, including its crew. The new labour monitoring practices we are seeing emerge today – with the stated aim of improving the working environment – appear just as outlandish. Take, for example, Canon's Beijing office, which has installed smart cameras that prevent any action from being performed (such as scheduling a meeting, accessing certain rooms, etc.) unless they detect a smile. In Europe, some companies are offering their employees the chance to participate in business-related trials which involve supplying them with glasses that establish emotion indicators. One example is the Shore app, developed by the Fraunhofer Institute for Integrated Circuits IIS in Germany, and which is used in Google's'smart glasses'.

As e-commerce goes mainstream, more product searches take place on Amazon every day than on Google or Facebook. The online economy is growing and growing but more evidence shows that Amazon's intricate engine, which now includes voice AI interfaces such as Echo, is threatening traditional retailers. The most recent study by digital marketing agency Wolfgang Digital has found that e-commerce transactions are up 32pc on last year and traffic is up 14pc. 'Amazon is now actively listening to you in your home so it can sell you more things next time you log on to Amazon on a screen. What's more, consumers are paying for the privilege of welcoming this 21st-century Trojan horse into their home' – ALAN COLEMAN It estimates that Ireland's online economy is worth €12.3bn, or 6pc of GDP in 2017.

Apple's Siri may not be the most capable voice assistant, or the most beloved. In the race to dominate the next generation of interfaces, though, Siri had one key advantage: a cushy home on hundreds of millions of iPhones. Now, however, Amazon has snuck Alexa onto iOS--making this look more and more like a blowout. Amazon's shopping app, already a top destination, now becomes a Trojan Horse for Amazon's most promising product in years. That puts iPhone and iPad owners just two taps away--one to open the Amazon app, the next to activate the microphone--from a voice assistant that doesn't just rival Siri, but surpasses it in significant ways.

It was Assistant, the artificially intelligent digital helper that caters to your every whim and powers your every interaction. Assistant is invisible, in the design-jargon sense. The omnipresent concierge works in the background, predicting your needs, processing your requests, and offering neatly parceled answers to your questions. You never see the cogs behind it, you merely type (or speak) a command and read (or hear) tailored responses served on screen or through a speaker. This requires more than a smartphone, which explains the gadgets Google announced Tuesday.

Stephanie Palermo wasn't interested in living in a "smart home" outfitted with web-connected appliances controlled remotely by phone or computer. She didn't need her fridge to have Wi-Fi or her blinds to close themselves. But when Amazon temporarily discounted the price on its voice-controlled Echo speaker to 99 for Amazon Prime members, there was "a low barrier to entry," and the 28-year-old from Belmont, Calif., was willing to take a risk. She started using the cylindrical device as a hands-free speaker. During board game nights, she'd tell Alexa -- the artificial intelligence that powers the Echo -- to play themed music from Pandora.

Shifting users toward intent-driven interactions is key to making AI work. Take this typical Spotify interaction: Open your phone, open Spotify, click to search, type what you want to hear. If you're just listening on your phone, you're done. Anything else takes a bit more work. "If I want to stream music to speakers in my living room, that's multiple steps I have to take, and I have to work through discovering the trigger points on the app," Mann says.

It was Assistant, the artificially intelligent digital helper that caters to your every whim and powers your every interaction. Assistant is invisible, in the design-jargon sense. The omnipresent concierge works in the background, predicting your needs, processing your requests, and offering neatly parceled answers to your questions. You never see the cogs behind it, you merely type (or speak) a command and read (or hear) tailored responses served on screen or through a speaker. This requires more than a smartphone, which explains the gadgets Google announced Tuesday.