> daily_signal(2026_07_08)

Meta's new Muse AI can put your face in someone else's AI-generated photo — it's on by default for every public Instagram account, and Meta says you'll never be notified when it happens.

PickBits Daily Signal · Wednesday, July 8, 2026

1. Meta's new Muse AI can put your face in someone else's AI-generated photo — it's on by default for every public Instagram account, and Meta says you'll never be notified when it happens.

Meta Superintelligence Labs launched Muse Image on July 7, 2026 — the company's first in-house AI image generator, live now in the Meta AI app, Instagram Stories, and WhatsApp. Tag a public Instagram account in a prompt and Muse can pull that person's photos to generate new images incorporating their likeness — no request, no consent, and per Meta's own help page, no notification, ever. The setting is opt-out, buried three menus deep in Instagram's 'Sharing and reuse' settings, and opting out only blocks future generations — anything already made using your face stays out there.

Key fact: If you have a public Instagram account, open Instagram now, tap the menu (three lines, top right), go to 'Sharing and reuse,' and toggle off both 'Posts' and 'Reels' under the AI/content-reuse permissions — this is on by default, meaning your face can already be used in someone else's AI image unless you turn it off yourself.

techcrunch.com · digitaltrends.com · primary source

2. The US just put a $10 million bounty on two Russian hacking crews who tricked government officials, journalists, and activists into handing over their Signal and WhatsApp backup keys.

The State Department's Rewards for Justice program announced on June 29, 2026 that it will pay up to $10 million for information on members of UNC4221 and UNC5792, hacking groups linked to Russia's FSB, Border Guard Service, and military intelligence. Their method wasn't a software exploit — it was social engineering: fake 'official support' texts, doctored Signal group-invite pages, and prompts to share verification codes or backup keys, giving attackers persistent access to encrypted messaging accounts even after a victim tried to start fresh.

Key fact: If you use Signal or WhatsApp for sensitive communication, never share a verification code, PIN, or backup/recovery key with anyone who contacts you first, even if the message looks like it's from Signal, WhatsApp, or your organization's IT/support team — that's the exact method this $10M-bounty campaign used.

therecord.media · primary source

3. Apollo's chief economist published a chart this week showing zero AI profit-margin lift at the 493 S&P 500 companies outside the 'Magnificent Seven' — his own read: Wall Street is pricing in gains that haven't shown up anywhere yet.

Torsten Slok, Apollo's chief economist, published a note and chart on July 7, 2026 laying out what Wall Street has priced into the roughly 493 non-'Magnificent Seven' S&P 500 companies: AI-driven margin expansion that, by his own count, isn't showing up yet. In regulated, process-heavy sectors — healthcare, banking, energy, pharma, manufacturing — the compliance and workflow overhauls needed to actually capture AI productivity gains take years, not months, and even where gains occur they're often too diffuse to show up cleanly on a balance sheet. Slok's own conclusion: if that timeline runs to 2029 instead of this year, a lot of AI-premised stock valuations face a hard correction.

Key fact: If you hold AI-adjacent stocks (directly or via index/retirement funds), check what fraction of your exposure assumes near-term margin expansion in non-tech sectors (healthcare, banking, energy, manufacturing) rather than the 'Magnificent Seven' — Slok's warning is specifically that this second bucket's AI payoff may be years away, not months.

the-decoder.com · apollo.com · ft.com · primary source

4. Meta's new brain-reading AI turns silent, intended typing into text — no surgery, no implant, just a cap that reads your brain's magnetic field, and the accuracy keeps climbing.

Meta FAIR's Brain2Qwerty v2, released July 2026, reconstructs full sentences from brain activity recorded entirely outside the skull — using magnetoencephalography (MEG) instead of an implant. In tests with nine healthy volunteers typing sentences they'd just heard, the model's average word error rate fell to 39% (from 55% without a language model), with the best participant reaching 22% and more than a quarter of sentences decoded perfectly. Surgical brain implants still do better — under 2% word error — but Meta's researchers say accuracy 'keeps climbing with more data, and no ceiling is in sight yet,' and portable, room-temperature MEG sensors are the path toward a device paralyzed patients who can't speak or move could actually use, without ever undergoing surgery.

Key fact: If you or someone you know lives with paralysis or a condition affecting speech/movement (ALS, brainstem stroke, locked-in syndrome), this remains research-stage — not yet a clinical device — but MEG-based non-invasive BCI is the more accessible track (no surgery, lower cost potential) worth watching as an alternative to implant-based options; ask a care team or university BCI research program about non-invasive trial eligibility as this moves toward the clinic.

the-decoder.com · github.com · nature.com · primary source