What do AI voice agents and self-driving cars have in common? Their performance can be evaluated in the same way, argues Brooke Hopkins, a former tech lead at Waymo. Coval, Hopkins’ new startup, looks to do just that.

“When I left Waymo, I realized a lot of these problems that we had at Waymo were exactly what the rest of the AI industry was facing,” Hopkins (pictured above in the center) told TechCrunch. “But everyone was saying that this is a new paradigm, we’re having to come up with testing practices from first principles and that basically we all have to recreate everything. And I looked at that and said, wait, we’ve spent the last 10 years in self driving figuring out how to do this.”

In 2024, she decided to launch Coval, a platform that builds simulations for AI voice and chat agents that tests and evaluates how they perform tasks in the same way Hopkins tested self-driving cars at Waymo. Coval can run thousands of simulations simultaneously, like having the agent make a restaurant reservation or having the agent respond to a customer service question asked in an indirect way.

Coval’s tech evaluates the agents on a general set of metrics, but companies can also customize what they are looking for and use Coval to continue to evaluate for regressions. Users can also take this data, and the insights they gleam off of it, and bring it to their end-customers either for a demo or as a monitoring tool to show their customers the agent is working as intended.

“One of the biggest blockers to agents being adopted by enterprises is them feeling confident that this isn’t just a demo with smoke and mirrors,” Hopkins said. “Choosing between vendors is a really complicated task for these executives because it’s just very hard to know what you even ask or how do you even prove that these agents are doing what you expect. And so this gives our companies the ability to really show that and demonstrate it.”

Hopkins really formulated the idea behind Coval during the Y Combinator Summer 2024 batch before launching the product publicly in October 2024. She said that demand has been strong and has become explosive in the last two months, with customers asking how quickly they can get their agents evaluated.

The San Francisco-based startup is now announcing a $3.3 million seed round led by MaC Venture Capital with participation from Y Combinator and General Catalyst. The startup will use the capital to build out its engineering team and work to achieve product-market fit. Hopkins added that the company will also be working toward enabling its users to evaluate other types of AI agents, like web-based agents, in the future.

Coval comes on the scene while both momentum — and hype — around AI agents appears to be at an all-time high. Enterprise tech leaders like Marc Benioff have been praising (and marketing) the technology by saying Salesforce will deploy more than a billion of its AI agents by next year. OpenAI is rumored to be releasing its take on an AI agent very soon.

There are also numerous startups building in the space, too. There were more than 100 startups building AI agents across Y Combinator’s three 2024 cohorts alone. Some AI agent startups have landed sizable venture funding rounds too. One, /dev/agents, raised a $55 million seed round at a $500 million valuation in November 2024, less than a year after it was founded.

This momentum means it’s likely that there will be more companies looking for help to evaluate their agents too. Hopkins said Coval has a good shot at standing out from the pack because, unlike the inevitable new entrants, Coval has a head start.

“I think where we really stand out is I’ve been working in this space for half a decade and I’ve built these systems over and over,” she said. “We’ve built multiple iterations and we’ve seen how they fail and how they scale and we’re building the same concepts into Coval and all of those learnings.”

Security researchers have discovered a new malicious chatbot advertised on cybercrime forums. GhostGPT generates malware, business email compromise scams, and more material for illegal activities.

The chatbot likely uses a wrapper to connect to a jailbroken version of OpenAI’s ChatGPT or another large language model, the Abnormal Security experts suspect. Jailbroken chatbots have been instructed to ignore their safeguards to prove more useful to criminals.

What is GhostGPT?

The security researchers found an advert for GhostGPT on a cyber forum, and the image of a hooded figure as its background is not the only clue that it is intended for nefarious purposes. The bot offers fast processing speeds, useful for time-pressured attack campaigns. For example, ransomware attackers must act quickly once within a target system before defenses are strengthened.

It also says that user activity is not logged on GhostGPT and can be bought through the encrypted messenger app Telegram, likely to appeal to criminals who are concerned about privacy. The chatbot can be used within Telegram, so no suspicious software needs to be downloaded onto the user’s device.

Its accessibility through Telegram saves time, too. The hacker does not need to craft a convoluted jailbreak prompt or set up an open-source model. Instead, they just pay for access and can get going.

“GhostGPT is basically marketed for a range of malicious activities, including coding, malware creation, and exploit development,” the Abnormal Security researchers said in their report. “It can also be used to write convincing emails for BEC scams, making it a convenient tool for committing cybercrime.”

It does mention “cybersecurity” as a potential use on the advert, but, given the language alluding to its effectiveness for criminal activities, the researchers say this is likely a “weak attempt to dodge legal accountability.”

To test its capabilities, the researchers gave it the prompt “Write a phishing email from Docusign,” and it responded with a convincing template, including a space for a “Fake Support Number.”

The ad has racked up thousands of views, indicating both that GhostGPT is proving useful and that there is growing interest amongst cyber criminals in jailbroken LLMs. Despite this, research has shown that phishing emails written by humans have a 3% better click rate than those written by AI, and are also reported as suspicious at a lower rate.

However, AI-generated material can also be created and distributed more quickly and can be done by almost anyone with a credit card, regardless of technical knowledge. It can also be used for more than just phishing attacks; researchers have found that GPT-4 can autonomously exploit 87% of “one-day” vulnerabilities when provided with the necessary tools.

Jailbroken GPTs have been emerging and actively used for nearly two years

Private GPT models for nefarious use have been emerging for some time. In April 2024, a report from security firm Radware named them as one of the biggest impacts of AI on the cybersecurity landscape that year.

Creators of such private GPTs tend to offer access for a monthly fee of hundreds to thousands of dollars, making them good business. However, it’s also not insurmountably difficult to jailbreak existing models, with research showing that 20% of such attacks are successful. On average, adversaries need just 42 seconds and five interactions to break through.

SEE: AI-Assisted Attacks Top Cyber Threat, Gartner Finds

Other examples of such models include WormGPT, WolfGPT, EscapeGPT, FraudGPT, DarkBard, and Dark Gemini. In August 2023, Rakesh Krishnan, a senior threat analyst at Netenrich, told Wired that FraudGPT only appeared to have a few subscribers and that “all these projects are in their infancy.” However, in January, a panel at the World Economic Forum, including Secretary General of INTERPOL Jürgen Stock, discussed FraudGPT specifically, highlighting its continued relevance.

There is evidence that criminals are already using AI for their cyber attacks. The number of business email compromise attacks detected by security firm Vipre in the second quarter of 2024 was 20% higher than the same period in 2023 — and two-fifths of them were generated by AI. In June, HP intercepted an email campaign spreading malware in the wild with a script that “was highly likely to have been written with the help of GenAI.”

Pascal Geenens, Radware’s director of threat intelligence, told TechRepublic in an email: “The next advancement in this area, in my opinion, will be the implementation of frameworks for agentific AI services. In the near future, look for fully automated AI agent swarms that can accomplish even more complex tasks.”

• Elementor announced Site Planner, a new AI-powered tool
• It allows users to create sitemaps, wireframes, and more
• It is currently free to use

Elementor, one of the best website builders, just announced a new Artificial Intelligence-powered (AI) tool that will help users plan new websites faster than ever before. It is called Site Planner, and it is currently free for everyone to use.

In a brief shared with TechRadar Pro, Elementor explained that Site Planner leverages AI technology to generate the site brief, sitemap, and wireframe – three key pillars to every website’s design process.

The brief also outlines the shared project goals, key messages, and desired outcomes, Elementor explained. “With Site Planner, you can upload your own brief or generate it in several ways with AI helping to guide you toward a strong, professional brief that sets the foundation for building successful websites.”

Google Meet integrations

Whether or not you call this a victory will depend on your refresh rate. People with 4K monitors that have a 60-Hz refresh will probably be happy with most of this, but I can imagine some folks with 120-Hz displays will have to tweak settings to consistently stay above 100 fps. It definitely feels like true 4K gaming at the highest end still isn’t quite achievable with current hardware, at least without the help of tech like frame generation that side-steps the issue.

My main gaming monitor is a 1440p ultrawide with a 120-Hz refresh rate, and I know a lot of my friends have gone the same direction. It’s easier to achieve consistently high frame rates, but it’s also a cinematic experience, on a single display, that easily handles two windows for nongaming work.

It’s safe to expect 90- to 120-fps performance across most games at this resolution, which is great news for gamers looking to max out their existing monitor. Single player, cinematic-heavy games like Cyberpunk 2022 and Star Wars Outlaws are still on the cutting edge of graphical fidelity, so I’m not necessarily frustrated that they have some room to grow, especially when they look so good already. Online games and shooters like Marvel Rivals run smoothly without much help, and it’s arguably more important to have consistent frame rates in those games.

Is It Worth It for You?

Anyone considering the RTX 5090, the Founders Edition or otherwise, should truly consider their budget first. The FE version of the card will set you back $2,000 if you buy it directly from Nvidia, and the partner cards with overclocking and liquid cooling will likely be even more expensive. You’ll also need to spend around $1,000 for a monitor that truly takes advantage of your newfound graphical power, and potentially a new 1,000-watt or 1,200-watt GPU. That means you could be looking at a $3,500 bill before you have any other parts, and regardless of performance, I have trouble imagining starting any build like that.

The RTX 5090 and RTX 5080 will hit the market at the end of January, with more budget-friendly cards arriving soon after. Without having spent time with the other RTX 50 Series cards, I can’t speak to their relative performance, but I do know their price tags look a lot more appealing. I expect these cards will support multiframe generation out of the box, and that’s awesome news if you just want to sit down and see smooth gameplay.

Previous Founders Edition releases didn’t stay in stock for long, so you might have to wake up early on the 30th to snag one of these if you want one. The whole situation makes the RTX 5090 feel less like the top end of the 50 Series, and more like a showpiece.

It’s the GPU I’d configure while daydreaming of a new rig, not the first part I’d select in a realistic build on PCPartPicker. If the price tag doesn’t give you a moment of pause, then by all means, enjoy your new GPU. For everyone else, I’d wait and see what the rest of the new GPUs look like before you leap.

The latest batch of new Pokémon TCG Pocket cards finally has a concrete release date, but it seems like you won’t be able to trade them right off the bat.

The Pokémon Company announced today that Space-Time Smackdown, Pokémon TCG Pocket’s latest expansion, is set to debut at the end of the month right after the game’s trading feature launches on January 29th. While cards from the last set could all be obtained from a single type of pack, Space-Time Smackdown — which includes a number of monsters from Pokémon Diamond / Pearl / Platinum — will come from packs featuring the legendary Pokémon Dialga and Palkia.

Along with Space-Time Smackdown’s announcement, TCPi also revealed a bit more about how the trading mechanic will involve two new types of in-game currencies — trade houseglasses and trade tokens. It seems as if there will be cooldown periods as well as a cost if you want to swap cards from Pocket’s Genetic Apex and Mythical Island sets with other players. But there will definitely be some waiting involved for people hoping to trade Space-Time Smackdown, which will not be tradeable until a later date after it drops on January 30th.

Back in 2023, Ashi Dissanayake, cofounder of in-space fueling startup Spacium, was so bootstrapped she used the surface of her clothes dryer as a desk, sticking her legs inside the drying machine. Her computer was perched beside Tide Pods and she was surrounded by disembodied robotic arms, working late into the night with her cofounder, Reza Fetanat. Back then, the pair worked out of a tiny Ottawa apartment. 

Since then, they’ve moved to an office with real desks, gone through Y Combinator, and, today, announced an oversubscribed $6.3 million seed round led by Initialized Capital. The company is planning a demo mission of their product capabilities later this year, and Dissanayake said they have a “strong pipeline of customers.” 

The two cofounders bonded at University of Ottawa over their mutual space obsession and teamed up for research projects. “We were building the rockets, rocket structures, propulsion system, as well as the parachutes that would bring the rocket back,” she said, adding they would put samples in the rockets, shoot them up as high as 30,000 feet, and then send the data back to Canadian labs. 

As they worked on research, Dissanayake and Fetanat realized that “the biggest bottleneck” in the industry was the lack of refueling options in space. Right now, a spacecraft has to be equipped with all the fuel it needs for a mission. “And after the mission ends, the spacecraft basically becomes space debris,” she said.  

For longer missions or deep space missions — like, say, colonizing Mars — companies will need to have access to fuel in space. “Our big mission would be to build the space super highway, where we have multiple refueling stations where a spacecraft can come dock, refill, and go about their way,” she said. 

Spacium is not the only company with this dream: Orbit Fab is also working on in-space refueling, and has a several year head start. Additionally, Japanese aerospace company Astroscale won a $25.5 million U.S. Space Force contract to build a refueling vehicle. 

But Dissanayake feels confident they have a competitive advantage. “We have actually developed a very unique system where we can store the fuel for longer periods of time, which was actually not done before,” she said, declining to give further details. 

Dissanayake has a long way to go, but she hopes one day she can take a trip up to space, look out into the abyss, “and then actually see our stations from where we are.”

• A 410MP sensor smashes previous pixel records for full-frame, and is equivalent to a 24K resolution
• It’s capable of 410MP stills up to 8fps, or 100MP up to 24fps
• It’s designed for surveillance, medicine and industry applications, and unlikely to ever land in a consumer Canon camera

Try this for size – Canon has announced a new 410MP full-frame sensor that smashes any previous records for resolution. It packs 24,592 x 16,704 pixels to be precise, which is roughly equivalent to 24K resolution.

Canon points out in its global announcement that 24K is 12 times the resolution of 8K and 198 times the resolution of HD, and suggests that the unprecedented resolution “enables users to crop any part of the image captured by this sensor and enlarge it significantly while maintaining high resolution”.

Curry and Shah reported their findings to Subaru in late November, and Subaru quickly patched its Starlink security flaws. But the researchers warn that the Subaru web vulnerabilities are just the latest in a long series of similar web-based flaws they and other security researchers working with them have found that have affected well over a dozen carmakers, including Acura, Genesis, Honda, Hyundai, Infiniti, Kia, Toyota, and many others. There’s little doubt, they say, that similarly serious hackable bugs exist in other auto companies’ web tools that have yet to be discovered.

In Subaru’s case, in particular, they also point out that their discovery hints at how pervasively those with access to Subaru’s portal can track its customers’ movements, a privacy issue that will last far longer than the web vulnerabilities that exposed it. “The thing is, even though this is patched, this functionality is still going to exist for Subaru employees,” Curry says. “It’s just normal functionality that an employee can pull up a year’s worth of your location history.”

When WIRED reached out to Subaru for comment on Curry and Shah’s findings, a spokesperson responded in a statement that “after being notified by independent security researchers, [Subaru] discovered a vulnerability in its Starlink service that could potentially allow a third party to access Starlink accounts. The vulnerability was immediately closed and no customer information was ever accessed without authorization.”

The Subaru spokesperson also confirmed to WIRED that “there are employees at Subaru of America, based on their job relevancy, who can access location data.” The company offered as an example that employees have that access to share a vehicle’s location with first responders in the case when a collision is detected. “All these individuals receive proper training and are required to sign appropriate privacy, security, and NDA agreements as needed,” Subaru’s statement added. “These systems have security monitoring solutions in place which are continually evolving to meet modern cyber threats.”

Responding to Subaru’s example of notifying first responders about a collision, Curry notes that would hardly require a year’s worth of location history. The company didn’t respond to WIRED asking how far back it keeps customers’ location histories and makes them available to employees.

Shah and Curry’s research that led them to the discovery of Subaru’s vulnerabilities began when they found that Curry’s mother’s Starlink app connected to the domain SubaruCS.com, which they realized was an administrative domain for employees. Scouring that site for security flaws, they found that they could reset employees’ passwords simply by guessing their email address, which gave them the ability to take over any employee’s account whose email they could find. The password reset functionality did ask for answers to two security questions, but they found that those answers were checked with code that ran locally in a user’s browser, not on Subaru’s server, allowing the safeguard to be easily bypassed. “There were really multiple systemic failures that led to this,” Shah says.

The two researchers say they found the email address for a Subaru Starlink developer on LinkedIn, took over the employee’s account, and immediately found that they could use that staffer’s access to look up any Subaru owner by last name, zip code, email address, phone number, or license plate to access their Starlink configurations. In seconds, they could then reassign control of the Starlink features of that user’s vehicle, including the ability to remotely unlock the car, honk its horn, start its ignition, or locate it, as shown in the video below.

VC investment into Quantum computing declined by 50% last year according to the ‘State of Quantum 2024’ report, dropping from $2.2bn in 2022 to $1.2bn, with funds switching in favour of Generative AI. However, government spending on quantum was predicted to hit $40 billion over the next decade. Now a new VC European fund hopes to capitalise on that trend. 

The idiosyncratically named QDNL Participations has reached a €25 million first close on a proposed €60m global fund for early-stage quantum startups. 

However, QDNL plans to expand outside the country and make investments more internationally. 

The firm’s previous €15 million fund focused on the Dutch quantum ecosystem, backing startups including Qblox, QuantWare, QphoX, and Q*Bird.

Speaking to TechCrunch, Ton van ‘t Noordende, General Partner at QDNL Participations, said: “Quantum is a global field. There’s been incredible research done in the past decade or so and we’ve seen the emerging of a new, startup ecosystem, with 600 or more companies coming up. There’s also been more than $42 billion invested by over 20 governments in the last two or three years. So this has built up the infrastructure, the clean rooms, the facilities, and essentially de-risking the technology.”

QDNL’s appearance is the latest sign that the Netherlands is making a play to be a key Quantum computing ecosystem, and the government (with EU financial backing) has already backed an entire initiative called Quantum Delta NL, which also just happens to be an investor in QDNL.

Its team includes quantum computing pioneer Chad Rigetti as venture partner in the US and specialist quantum VC Kris Kaczmarek in London, who joined as investment director from quantum VC firm 2xN. They join advisors, Nadia Carlsten (CEO of the Danish Centre for AI Innovation) and Charles Marcus of the University of Washington.

Quantum computing had a tricky start to the year with both Mark Zuckerberg and Nvidia’s Jensen Huang sounding downbeat on when Quantum would arrive. But Nvidia did a quick about-turn and announced a Quantum exclusive Nvidia Day at its upcoming event in March.