Conversation with Apple Senior Vice President of Software Engineering Craig Federighi: The best safety is like an airbag

In order to improve service quality, this call will be recorded.

Everyone who has called corporate customer service must be familiar with this sentence. But "If you don't agree to the recording, please press 1." You will only hear this from Apple's customer service phone.

Among the many new features in iOS 17, the privacy upgrade seems to be the most inconspicuous one, but within Apple, it is regarded as an unshakable creed.

After this fall’s event, I had a brief conversation with Craig Federighi, Apple’s senior vice president of software engineering. While the conversation centered around privacy and security, the range of products involved was unexpected.

Craig focuses on software engineering development, but he believes that security is not just about the software level.

Only by solving problems in an integrated way at every level, from the chip all the way to the software, can you create the most secure products on the market.

Apple has set up a team dedicated to privacy and security, and Craig said the team will design around privacy and security from the beginning, rather than waiting until everything is ready and adding it at the end. Because many of the key points of privacy issues cannot be solved by hardware engineers and software engineers alone.

For example, the earliest TouchID built the first layer of protection for the iPhone through fingerprint recognition. In fact, all fingerprint recognition information is only stored in a subsystem on the SoC chip.

This method is called "security compartment" by Craig. It separates various security subsystems through componentization and isolates them separately in hardware and software. Even if an attacker can break through one of the layers, he cannot Enter other levels.

This is like a bank with layers of protection. You must first take the elevator down to the 100th floor underground and pass through multiple layers of isolation before you can enter the vault.

Unlike other giants that are building data centers, Apple prefers to deploy its technology locally and give priority to the computing power of local chips.

▲ Apple achieves input prediction through optimization of the device-side Transformer model

Efficiency and cost are naturally on the one hand, but safety is a more important consideration.

For example, ChatGPT, which became popular this year, has almost eaten up the computing resources of Microsoft's Azure cloud computing center, because hundreds of millions of questions around the world have to be transmitted and deduced through network servers.

When it comes to security, if you can only trust one person, it is you, not a commercial company that advertises security but leaves backdoors in lengthy user agreements.

"What you say to Siri is actually not related to your Apple ID." The privacy team told me that the audio never leaves the device. When the user activates Siri, the device system will generate a random identity. The identification code is managed separately, and all instructions are only run on the device side.

In the same way, this can also explain why when you get a new iPhone, you always have to enter your fingerprint or facial information again.

This persistence has also fed back into Apple's core manufacturing business. In the past two years, Neural Engine, like CPU and GPU, has been regarded as the most important performance indicator for A-series and M-series chips.

For Apple, the performance of the chip is not only how high the running score is and how smooth the experience is, but also whether it can build a sufficiently secure wall.

As for some indispensable information, Apple will follow the principle of "data minimization":

For example, when a user searches for a geographical location, the map app will use a "blurring" method: after 24 hours, the precise location searched by the user will be blurred into a general area.

In the early years, Apple also introduced an algorithm called "Differential Privacy" to ensure security. This algorithm scrambles personal privacy data and mixes it with millions of other data. Even Apple itself cannot derive data about specific individuals, but the overall data can help Apple find the best data. Popular emojis, best quick typing suggestions.

To this day, iOS is the only system that has never experienced a large-scale malware attack.

Craig said that Apple will not let users give up their privacy for the sake of experience, and conversely, Apple will not set up obstacles for security. Therefore, the team adopted various methods to make security protection invisible.

Just like the airbags in your car, you may not know how they work, you may not even know they exist, but they are there to protect you. I think this is the best safety.

The business model of today's Internet giants is basically based on the exchange of benefits: you hand over your personal information in exchange for personalized services. The private information collected without your knowledge not only allows commercial companies to predict your future behavior, but also makes you a target for targeted advertising.

Harari, author of A Brief History of Humanity, described the loss of privacy this way:

It's like the indigenous peoples of Africa and America who inadvertently sold their entire country to a European country in exchange for beads of various colors and cheap jewelry.

A few years ago, the Cambridge Analytica scandal exposed the consequences of ignoring "privacy" to the world: any company with hundreds of millions of users can have the power to affect the world if its data is abused.

Computer scientist Jaron Lanier blames the negative side of technology and algorithms on free business models.

As computers get smarter and algorithms become more precise, and everything works to get you to see more ads, everything seems to be going haywire. I don’t call them social networks anymore, I now call them behavioral modification empires. .

As a giant technology company, Apple’s uniqueness lies in the fact that it is a company that relies on hardware sales. Its god is the general public, not advertisers.

But doing security well does not come from slogans and claims. It is about the art of combining software and hardware, about the control of core components and operating systems, and more about the understanding of the human nature behind the technology.

Love News Feed and Menu, and am trying to process more information than fat intake every day.

# Welcome to follow the official WeChat public account of Aifaner: Aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo