While many existing risks and controls can apply to generative AI, the groundbreaking technology has many nuances that require new tactics, as well. 

Models are susceptible to hallucinations, or the production of inaccurate content. Other risks include the leaking of sensitive data via a model’s output, tainting of models that can allow for prompt manipulation and biases as a consequence of poor training data selection or insufficiently well-controlled fine-tuning and training. 

Ultimately, conventional cyber detection and response needs to be expanded to monitor for AI abuses — and AI should conversely be used for defensive advantage, said Phil Venables, CISO of Google Cloud.

“The secure, safe and trusted use of AI encompasses a set of techniques that many teams have not historically brought together,” Venables noted in a virtual session at the recent Cloud Security Alliance Global AI Symposium.

Lessons learned at Google Cloud

Venables argued for the importance of delivering controls and common frameworks so that every AI instance or deployment does not start all over again from scratch. 

“Remember that the problem is an end-to-end business process or mission objective, not just a technical problem in the environment,” he said. 

Nearly everyone by now is familiar with many of the risks associated with the potential abuse of training data and fine-tuned data. “Mitigating the risks of data poisoning is vital, as is ensuring the appropriateness of the data for other risks,” said Venables. 

Importantly, enterprises should ensure that data used for training and tuning is sanitized and protected and that the lineage or provenance of that data is maintained with “strong integrity.”

“Now, obviously, you can’t just wish this were true,” Venables acknowledged. “You have to actually do the work to curate and track the use of data.”

This requires implementing specific controls and tools with security built in that act together to deliver model training, fine-tuning and testing. This is particularly important to assure that models are not tampered with, either in the software, the weights or any of their other parameters, Venables noted. 

“If we don’t take care of this, we expose ourselves to multiple different flavors of backdoor risks that can compromise the security and safety of the deployed business or mission process,” he said. 

Filtering to fight against prompt injection

Another big issue is model abuse from outsiders. Models may be tainted through training data or other parameters that get them to behave against broader controls, said Venables. This could include adversarial tactics such as prompt manipulation and subversion. 

Venables pointed out that there are plenty of examples of people manipulating prompts both directly and indirectly to cause unintended outcomes in the face of “naively defended, or flat-out unprotected models.” 

This could be text embedded in images or other inputs in single or multimodal models, with problematic prompts “perturbing the output.”

“Much of the headline-grabbing attention is triggering on unsafe content generation, some of this can be quite amusing,” said Venables.

It’s important to ensure that inputs are filtered for a range of trust, safety and security goals, he said. This should include “pervasive logging” and observability, as well as strong access control controls that are maintained on models, code, data and test data, as well. 

“The test data can influence model behavior in interesting and potentially risky ways,” said Venables. 

Controlling the output, as well

Users getting models to misbehave is indicative of the need to manage not just the input, but the output, as well, Venables pointed out. Enterprises can create filters and outbound controls — or “circuit breakers” —around how a model can manipulate data, or actuate physical processes. 

“It’s not just adversarial-driven behavior, but also accidental model behavior,” said Venables. 

Organizations should monitor for and address software vulnerabilities in the supporting infrastructure itself, Venables advised. End-to-end platforms can control the data and the software lifecycle and help manage the operational risk of AI integration into business and mission-critical processes and applications. 

“Ultimately here it’s about mitigating the operational risks of the actions of the model’s output, in essence, to control the agent behavior, to provide defensive depth of unintended actions,” said Venables. 

He recommended sandboxing and enforcing the least privilege for all AI applications. Models should be governed and protected and tightly shielded through independent monitoring API filters or constructs to validate and regulate behavior. Applications should also be run in lockdown loads and enterprises need to focus on observability and logging actions. 

In the end, “it’s all about sanitizing, protecting, governing your training, tuning and test data. It’s about enforcing strong access controls on the models, the data, the software and the deployed infrastructure. It’s about filtering inputs and outputs to and from those models, then finally making sure you’re sandboxing more use and applications in some risk and control framework that provides defense in depth.”

Buying a home has always been complicated. You have to figure out how much money to put down and how that down payment will affect a monthly mortgage bill. Then there are the closing costs and fees. Kevin Bennett launched Further to try to help make the financial process easier to navigate — especially for first-time buyers.

Further is a fintech platform that walks users through the financial side of home buying. The company’s first product, which goes live Friday, is a calculator that shows what people can afford and what their monthly mortgage payments and closing costs could look like, among other metrics based on real-time interest rates.

Unlike other mortgage calculators that you can find on Zillow and LendingTree, Further looks to give users more than the numbers. It tells users how easy it will be for them to find a loan based on their financial status, whether they should wait to buy, or if they should pursue specific types of loans based on their financial profile, among others.

The platform is currently free to use. The company plans to monetize once it releases more product developments but declined to share details.

“A generation ago, our parents bought a $200,000 home with a 20% mortgage, and it was very straightforward,” Bennett said. “There was one kind of mortgage, and that’s what you did and it’s just more complicated. There are lots of kinds of mortgages. There are lots of implications. Homes are much more expensive now, so there’s just a lot more complexity, and it’s a much bigger financial decision.”

Last year Bennett found himself looking for something new to work on after stepping back from Caribou, the auto loan refinancing startup he co-founded in 2016 and where he served as CEO. He knew he wanted to do something else mission-oriented but wasn’t sure where.

He started looking into real estate, a category he said he’s always been fascinated with. The fact that his whole family works in real estate helped, too. He started talking to folks who had purchased their home within the last two years and found a lot of common pain points: People didn’t understand the process and were relying on homemade spreadsheets to try to figure out what they could afford.

Bennett also had a personal experience: He bought and sold a townhouse in his 20s and was surprised to find out he endured a $30,000 loss, despite selling the home for the original purchase price. That’s because he missed out on certain home improvements that could’ve increased the house’s value.

“You can’t hit the undo button once you buy that house,” Bennett said. “It felt like there was a gap in the market. It felt like it was a lot more complicated than it was a generation ago.”

He reached out to his friend Chris Baker, a real estate expert, and former head of product at EasyKnock, about his idea last year. The pair got to work fast. Their first conversation was November 3, 2023. They decided to work together in January, launched the product in April, and raised an undisclosed pre-seed round in June. Now, they are coming out of stealth.

“Our goal is to take care of the complicated jargon and stuff and really help you understand as easily as possible what it is you need to know, with transparency, obviously, but also putting you in the driver’s seat and in control,” he said.

The company’s previously undisclosed pre-seed round raised $4.1 million from investors including Link Ventures, Vesta Ventures, and Fidi Ventures, among others. Bennett said that fundraising wasn’t too challenging, as half of the capital the company raised was from investors who backed him while he was at Caribou. Bennett thinks his track record as a founder made a big difference. The company built its cap table intentionally to include angel investors who have experience in the real estate market, he said.

This kind of financial information and guidance seems like something a Zillow or Redfin would be ripe to copy especially considering Zillow already offers a mortgage calculator and some advice of its own. But Bennett said he wasn’t super concerned about the competition. He said he thinks that many companies either fall on the proptech side or the fintech side and rarely in the middle, as Further does, which gives it more of a moat.

But Further is definitely not the only company that sits between proptech and fintech that is aimed at consumers. Online mortgage startup Better.com, which allows consumers to browse for mortgage options or refinance an existing one, is a good example.

It will likely depend on what Further unveils in its planned Q1 product release that will include more features and capabilities, but Bennett didn’t share too many details just yet. For now, users can use Further to get an idea of what they can afford and what they can expect to pay when buying a house.

“My hope is that we can enable people with the right insights and information to make good decisions and plan for this really big part of their life in a way that gives them confidence, puts them at ease and and lets them focus on, you know, what they really want to focus on, which is kind of that that dream of being a homeowner,” Bennett said.

A newly proposed cosmic speed limit may constrain how fast anything in the universe can grow. Its existence follows from Alan Turing’s pioneering work on theoretical computer science, which opens the intriguing possibility that the structure of the universe is fundamentally linked to nature of computation.

Cosmic limits are not a new idea. While studying the relationship between space and time, Albert Einstein showed that nothing in the universe can exceed the speed of light, as part of his special theory of relativity. Now, Toby Ord at…

The addition of a 4.3-inch color TFT screen makes the new Wyze Scale Ultra one of the brand’s most expensive smart scales to date, but at $43.99, it’s still considerably cheaper than offerings from companies like Withings. It’s available from Wyze directly or from Amazon in white or black.

The Wyze Scale Ultra says it can track 13 different health metrics, including your heart rate, your metabolic age (a comparison of how your body burns calories at rest to others your age), and measurements of fat, muscle, and water.

Previous versions of Wyze’s smart scales featured simple segmented LED displays to display basic information like weight, BMI, and muscle mass, leaving more detailed breakdowns of your health metrics for an accompanying mobile app. The Wyze Scale Ultra can display more data, including how measurements like weight or body fat have fluctuated over time, and it’s customizable, so it only displays what you want it to.

Like the Wyze Scale X introduced in 2022, the Scale Ultra offers modes for easily weighing pets, babies, or luggage and a pregnancy mode that turns off the weak electrical current used for bioelectric impedance analysis (BIA) as an added safety precaution.

Connectivity includes both Bluetooth and Wi-Fi, and the Wyze Scale Ultra can automatically recognize and sync measured health metrics for up to eight different users — either to its mobile app or to the Apple Health, Google Fit, and Fitbit platforms. It’s not rechargeable, however. It runs on four AA batteries, which Wyze says will keep the scale powered for up to nine months.

LiteSpeed Cache, an immensely popular WordPress plugin for site performance optimization, suffered from a vulnerability which allowed threat actors to gain admin status.

With such elevated privileges, they would be able to perform all sorts of malicious activities on the compromised websites.