AI & Email Immutability: Concerns and Solutions

To ensure responsible AI usage, staff should grasp existing tools and employ them securely. Any tool use, including AI, must adhere to privacy-preserving principles. This includes data protection, following confidentiality and privacy policies.
Vendors are increasingly using AI for tasks like translation, categorization, and abuse detection, and measures are in place to ensure customer protection.

AI Accountability

AI-generated content requires human review and must undergo regular processes before use.

Bias Awareness

Actively seek out biases or hallucinations in AI output.

Human Authority

Ensure a way to appeal automated decisions to a human.

While plain text remains unchangeable, modern emails often incorporate remote content. Opening an email retrieves assets directly from the sender, which are cached temporarily and not permanently stored. If these assets change or disappear, the email provider can do little. As Google notes about AMP messages, the content can dynamically change over time.

A benefit is that senders can't use emails as read receipts, as providers can state they perform operations for immutability, similar to how other email operations proxy requests for privacy.

While it can be interesting to see an email update to show tracking progress, a new email could just be sent instead.

Plaintext versions are useful for content testing by displaying full message contents, although not often read outside the company. While some appreciate plaintext emails, it's not the main customer base.

Email formatting mechanics are simple and haven't changed much in over 50 years, making it straightforward until complexity is added. HTML email is flawed because content can be anywhere.

Text-only email remains important and widely used. Apple's private loading feature provides a fix: the mail server saves referenced content, ensuring what was served when the message was sent is always known.

Choosing an email provider isn't impacted by concerns about visual alterations to emails.

Cameras used for news documentation will likely require watermarks, fingerprinting, and authentication, similar to what Canon and Nikon are doing. This may seem like a gimmick now, but in the future, visuals from companies with these measures may be the only trustworthy ones.

Social media apps might encourage direct camera-to-post for unaltered documentation of reality. For example, filming a protest through Instagram could result in a "this is real" badge, while uploaded videos might be treated as potentially AI-generated.

Sophisticated AI could detect if a monitor is being recorded. Manipulating media is easier now, but courts have always had to deal with manipulated evidence. Lawyers must carefully select jurors, being mindful of their potential susceptibility to AI manipulation. Watermarking and social media platforms could also use watermarks.

Ordinary users don't care about mainstream media as much, which poses a bigger threat than AI videos. War involves actual deaths, and media manipulation can instigate it.

Submissions on platforms like Reddit/Twitter commonly include images with text or screenshots of headlines, which are often taken seriously without verification. Requiring photos to be signed by camera hardware might not prevent the spread of misinformation if images of text are enough for people to believe them.

This concentrates potential for misuse to the platform itself.

Audiences need to learn new ways to determine trustworthiness, since technology alone is not the answer. Some place more hope in social contracts instead of people. LIDAR sensors could be useful for verifying depth information in images.

Instagram, owned by someone keen on AI, doesn't seem likely to adopt the use of LIDAR.

Cameras can record screens showing AI-generated videos and still mark them as authentic. While screen pixels not perfectly mapping to image sensor pixels may be detectable now, screens will likely exist that can overcome this if elections depend on it.

Using LIDAR to prevent screen recording can be circumvented with a mirror setup. People need to take personal responsibility, because AI is not only for action shots.

People need to take personal responsibility and not blindly believe everything they see or hear, even from figures of authority, and verify claims before acting.

It is already possible to manipulate images with techniques used in the production of shows like The Mandalorian. They utilize LED walls and dynamic digital sets, with the ability to react to and manipulate digital content during live production, allowing entire CGI backgrounds to be manipulated in real-time.

Proving an email's authenticity is challenging. It requires retaining DKIM keys to check historical signatures to confirm the message hasn't been altered or forged. It is uncertain whether DKIM signing keys are issued in a Certificate Transparency log to verify key existence for a domain.

Until recently, email was sent and received without encryption, like a postcard, without security or privacy. A stated viewpoint is that an email is your copy and the sender cannot revise it. To be truly sure that no one is tampering with your email, do not share it with anyone else.

One user said that if their email provider adds a single AI feature they will return to self hosting.

Domain age, IP reputation, and compliance with DKIM/DMARC are important. Switching to hosted email solved delivery problems. Also, if people email you first from their domain and you reply, you are less likely to be blocked.

A user with a long-registered domain and a Class C IP range found Gmail started ignoring their email server. Before sending email, using mail-tester.com helps check DKIM setup and IP blacklisting.

Google's AI overview is better and faster than some proprietary AI assistants.

One user wanted to add users to their billing plan to give someone their own Fastmail Inbox and login. It makes sense for heavy email users but not so much for family members. They hoped there was a 5 accounts for the price of 3 thing like Spotify et al do.

One user with 16.5GB of storage usage found that upgrading to Gemini gave them 2TB of space, while their usage dropped to 12GB. That is disgusting. That is why the user might as well switch to fastmail. For another user, if you count for automatically categorized Bayesian spam, email is about 99% noise. Current AI could be used to enhance privacy and filter advertising.